diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000000000000000000000000000000000000..3bfd1681ac9775df50596503b80a315d1493bd8f
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,3 @@
+mathematics_dataset/mathematics_dataset/*/__pycache__/
+mathematics_dataset/mathematics_dataset/__pycache__/
+mathematics_dataset/mathematics_dataset/.venv/
\ No newline at end of file
diff --git a/README.md b/README.md
index 32897cd3e640101ba184f8c4ccd896981de3804a..9bc60a20eb26e6a868dcbe9712abfef898f36735 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,20 @@
 ---
 license: mit
 ---
+
+
+To run generation code within 'mathematics_dataset\mathematics_dataset\':
+- Activate python venv ``` .\.venv\Scripts\activate ```
+    - Requirements defined in requires.txt
+- Run ```python generate_to_file.py --output_dir ds```  to generate dataset to directory \ds
+
+Had to change enconding when opening files to utf-8 so that some characters are allowed (ã õ é)
+
+To obtain dataset with the correct amount of rows:
+- python ```generate_to_file.py --output_dir ds --per_train_module 1999998 --per_test_module 10000```
+
+This dataset creates train set (train-easy,train-medium,train-hard) and the creates extrapolation ("measure generalization along various axes of difficulty to beyond that seen during training") and interpolation ("test questions are distinct
+from the train questions") tests.
+On HugginFace only interpolate tests are used as test set.
+
+Some tests will not work, since they rely on english terms.
\ No newline at end of file
diff --git a/Translation_rules.xlsx b/Translation_rules.xlsx
new file mode 100644
index 0000000000000000000000000000000000000000..d1555389ca827e7eb365b1a6efc834eb9e3f1166
Binary files /dev/null and b/Translation_rules.xlsx differ
diff --git a/mathematics_dataset/CONTRIBUTING.md b/mathematics_dataset/CONTRIBUTING.md
new file mode 100644
index 0000000000000000000000000000000000000000..db177d4ac70f76c2172ce65ca33c830d6cbd3ac5
--- /dev/null
+++ b/mathematics_dataset/CONTRIBUTING.md
@@ -0,0 +1,28 @@
+# How to Contribute
+
+We'd love to accept your patches and contributions to this project. There are
+just a few small guidelines you need to follow.
+
+## Contributor License Agreement
+
+Contributions to this project must be accompanied by a Contributor License
+Agreement. You (or your employer) retain the copyright to your contribution;
+this simply gives us permission to use and redistribute your contributions as
+part of the project. Head over to <https://cla.developers.google.com/> to see
+your current agreements on file or to sign a new one.
+
+You generally only need to submit a CLA once, so if you've already submitted one
+(even if it was for a different project), you probably don't need to do it
+again.
+
+## Code reviews
+
+All submissions, including submissions by project members, require review. We
+use GitHub pull requests for this purpose. Consult
+[GitHub Help](https://help.github.com/articles/about-pull-requests/) for more
+information on using pull requests.
+
+## Community Guidelines
+
+This project follows
+[Google's Open Source Community Guidelines](https://opensource.google.com/conduct/).
diff --git a/mathematics_dataset/LICENSE b/mathematics_dataset/LICENSE
new file mode 100644
index 0000000000000000000000000000000000000000..d645695673349e3947e8e5ae42332d0ac3164cd7
--- /dev/null
+++ b/mathematics_dataset/LICENSE
@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/mathematics_dataset/README.md b/mathematics_dataset/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..d12161e88da78b17985457f1ff92da02c52a822e
--- /dev/null
+++ b/mathematics_dataset/README.md
@@ -0,0 +1,177 @@
+# Mathematics Dataset
+
+This dataset code generates mathematical question and answer pairs, from a range
+of question types at roughly school-level difficulty. This is designed to test
+the mathematical learning and algebraic reasoning skills of learning models.
+
+Original paper: [Analysing Mathematical
+Reasoning Abilities of Neural Models](https://openreview.net/pdf?id=H1gR5iR5FX)
+(Saxton, Grefenstette, Hill, Kohli).
+
+## Example questions
+
+```
+Question: Solve -42*r + 27*c = -1167 and 130*r + 4*c = 372 for r.
+Answer: 4
+
+Question: Calculate -841880142.544 + 411127.
+Answer: -841469015.544
+
+Question: Let x(g) = 9*g + 1. Let q(c) = 2*c + 1. Let f(i) = 3*i - 39. Let w(j) = q(x(j)). Calculate f(w(a)).
+Answer: 54*a - 30
+
+Question: Let e(l) = l - 6. Is 2 a factor of both e(9) and 2?
+Answer: False
+
+Question: Let u(n) = -n**3 - n**2. Let e(c) = -2*c**3 + c. Let l(j) = -118*e(j) + 54*u(j). What is the derivative of l(a)?
+Answer: 546*a**2 - 108*a - 118
+
+Question: Three letters picked without replacement from qqqkkklkqkkk. Give prob of sequence qql.
+Answer: 1/110
+```
+
+## Pre-generated data
+
+[Pre-generated files](https://console.cloud.google.com/storage/browser/mathematics-dataset)
+
+### Version 1.0
+
+This is the version released with the original paper. It contains 2 million
+(question, answer) pairs per module, with questions limited to 160 characters in
+length, and answers to 30 characters in length. Note the training data for each
+question type is split into "train-easy", "train-medium", and "train-hard". This
+allows training models via a curriculum. The data can also be mixed together
+uniformly from these training datasets to obtain the results reported in the
+paper. Categories:
+
+* **algebra** (linear equations, polynomial roots, sequences)
+* **arithmetic** (pairwise operations and mixed expressions, surds)
+* **calculus** (differentiation)
+* **comparison** (closest numbers, pairwise comparisons, sorting)
+* **measurement** (conversion, working with time)
+* **numbers** (base conversion, remainders, common divisors and multiples,
+  primality, place value, rounding numbers)
+* **polynomials** (addition, simplification, composition, evaluating, expansion)
+* **probability** (sampling without replacement)
+
+## Getting the source
+
+### PyPI
+
+The easiest way to get the source is to use pip:
+
+```shell
+$ pip install mathematics_dataset
+```
+
+### From GitHub
+
+Alternately you can get the source by cloning the mathematics_dataset
+repository:
+
+```shell
+$ git clone https://github.com/deepmind/mathematics_dataset
+$ pip install --upgrade mathematics_dataset/
+```
+
+## Generating examples
+
+Generated examples can be printed to stdout via the `generate` script. For
+example:
+
+```shell
+python -m mathematics_dataset.generate --filter=linear_1d
+```
+
+will generate example (question, answer) pairs for solving linear equations in
+one variable.
+
+We've also included `generate_to_file.py` as an example of how to write the
+generated examples to text files. You can use this directly, or adapt it for
+your generation and training needs.
+
+## Dataset Metadata
+The following table is necessary for this dataset to be indexed by search
+engines such as <a href="https://g.co/datasetsearch">Google Dataset Search</a>.
+<div itemscope itemtype="http://schema.org/Dataset">
+<table>
+  <tr>
+    <th>property</th>
+    <th>value</th>
+  </tr>
+  <tr>
+    <td>name</td>
+    <td><code itemprop="name">Mathematics Dataset</code></td>
+  </tr>
+  <tr>
+    <td>url</td>
+    <td><code itemprop="url">https://github.com/deepmind/mathematics_dataset</code></td>
+  </tr>
+  <tr>
+    <td>sameAs</td>
+    <td><code itemprop="sameAs">https://github.com/deepmind/mathematics_dataset</code></td>
+  </tr>
+  <tr>
+    <td>description</td>
+    <td><code itemprop="description">This dataset consists of mathematical question and answer pairs, from a range
+of question types at roughly school-level difficulty. This is designed to test
+the mathematical learning and algebraic reasoning skills of learning models.\n
+\n
+## Example questions\n
+\n
+```\n
+Question: Solve -42*r + 27*c = -1167 and 130*r + 4*c = 372 for r.\n
+Answer: 4\n
+\n
+Question: Calculate -841880142.544 + 411127.\n
+Answer: -841469015.544\n
+\n
+Question: Let x(g) = 9*g + 1. Let q(c) = 2*c + 1. Let f(i) = 3*i - 39. Let w(j) = q(x(j)). Calculate f(w(a)).\n
+Answer: 54*a - 30\n
+```\n
+\n
+It contains 2 million
+(question, answer) pairs per module, with questions limited to 160 characters in
+length, and answers to 30 characters in length. Note the training data for each
+question type is split into "train-easy", "train-medium", and "train-hard". This
+allows training models via a curriculum. The data can also be mixed together
+uniformly from these training datasets to obtain the results reported in the
+paper. Categories:\n
+\n
+* **algebra** (linear equations, polynomial roots, sequences)\n
+* **arithmetic** (pairwise operations and mixed expressions, surds)\n
+* **calculus** (differentiation)\n
+* **comparison** (closest numbers, pairwise comparisons, sorting)\n
+* **measurement** (conversion, working with time)\n
+* **numbers** (base conversion, remainders, common divisors and multiples,\n
+  primality, place value, rounding numbers)\n
+* **polynomials** (addition, simplification, composition, evaluating, expansion)\n
+* **probability** (sampling without replacement)</code></td>
+  </tr>
+  <tr>
+    <td>provider</td>
+    <td>
+      <div itemscope itemtype="http://schema.org/Organization" itemprop="provider">
+        <table>
+          <tr>
+            <th>property</th>
+            <th>value</th>
+          </tr>
+          <tr>
+            <td>name</td>
+            <td><code itemprop="name">DeepMind</code></td>
+          </tr>
+          <tr>
+            <td>sameAs</td>
+            <td><code itemprop="sameAs">https://en.wikipedia.org/wiki/DeepMind</code></td>
+          </tr>
+        </table>
+      </div>
+    </td>
+  </tr>
+  <tr>
+    <td>citation</td>
+    <td><code itemprop="citation">https://identifiers.org/arxiv:1904.01557</code></td>
+  </tr>
+</table>
+</div>
diff --git a/mathematics_dataset/mathematics_dataset/__init__.py b/mathematics_dataset/mathematics_dataset/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..d780829a1f5101333577a1ca9adcb960529213f1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/__init__.py
@@ -0,0 +1,13 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/algebra__polynomial_roots_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/algebra__polynomial_roots_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..350d9690f4c25f759b9007db7e0604f9e36153c2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/algebra__polynomial_roots_big.txt
@@ -0,0 +1,20 @@
+Seja 3*i**2/2 - 474313521*i/2 + 2134410723 = 0. Qual é i?
+9, 158104498
+Factoriza 3*g**2 - 169065244338*g + 2381921403588970088187.
+3*(g - 28177540723)**2
+Factoriza 3*x**3 + 33899714526*x**2 + 95765887044791569188*x - 191531774225181996504.
+3*(x - 2)*(x + 5649952422)**2
+Seja 3*v**4/4 + 9028809*v**3/2 + 1914074427*v**2/4 + 943494003*v + 469489800 = 0. Qual é v?
+-6019100, -104, -1
+Encontra i tal que -2*i**3/13 - 281030*i**2/13 - 434648094*i/13 - 1301415066/13 = 0.
+-138951, -1561, -3
+Resolve z**3/2 + 664681*z**2 + 221744124000*z + 560260252092672 = 0.
+-663408, -2546
+Supõe -t**3/5 - 1058377*t**2/5 + 234356101*t - 8150602719/5 = 0. Qual é t?
+-1059483, 7, 1099
+Supõe -2*i**2/9 + 100399784548*i/9 - 1260014592160602445538/9 = 0. Qual é i?
+25099946137
+Seja -o**4/3 - 296*o**3 + 149338560*o**2 - 1173502213632*o = 0. Calcula o.
+-24744, 0, 11928
+Supõe 3*a**2/4 - 19005912951*a/2 = 0. Calcula a?
+0, 12670608634
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_or_sub_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_or_sub_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5a2cb8c5cf22509b9ed2c8525dcb5cef0cd832aa
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_or_sub_big.txt
@@ -0,0 +1,20 @@
+Quanto é -41.924 retirado de -33604115882?
+-33604115840.076
+-22757918145374.91 + 0.5
+-22757918145374.41
+Determina 1.19 + -278715.37375089.
+-278714.18375089
+Quanto é -2.497416 retirado de -69846.3671?
+-69843.869684
+Quanto é 233.1458582 + -92.150525?
+140.9953332
+Quanto é 2 mais 1383039726125.669?
+1383039726127.669
+Quanto é -434368 mais -80.387315806?
+-434448.387315806
+Calcula 153631.982 - 2784044885.
+-2783891253.018
+Quanto é 92326064.2 - 1586?
+92324478.2
+Qual é a distância entre 4 e -263788773099.330361?
+263788773103.330361
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_sub_multiple_longer.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_sub_multiple_longer.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a93ca039f93886a8fe2c70b975f329c254f08232
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__add_sub_multiple_longer.txt
@@ -0,0 +1,20 @@
+Avalia (-17 - -34 - (-10 + (-3 - -10) - -9)) + 253 - ((17 - 9) + -5).
+261
+5 + 11 - (-67 - -98) - -10 - (0 + (2 - 0) + -3 - (36 + -93))
+-61
+Calcula (-1 + -186 - -2) + (-11756 - -11792) + -24 + 2 + -2 - (6 + (-2 - 1)).
+-176
+Quanto é (2 + -18 + 13 - 4) + 16 + 352 + (0 - (0 - -3)) + -4 + -10?
+344
+Avalia ((-54 - -140) + -133 - 48) + -11 + 2 + (2 + (-2 - -1) - 1).
+-104
+Calcula (8 - -1) + -27 + (25 - (-3 - (-16 - -3))) + 15 + 236 + 6 + 0.
+254
+Quanto é 9 - (3 - (-15 + (-260 - (-81 - -12)) + 1 + (-4 - -2 - 10)))?
+-211
+-8 + (284 - (-20 + (17 - 7))) - (-4 + (-5 - -5)) - (0 + 9)
+281
+Calcula 0 - (-301 + 3) - (1 + (-30 - (8 + -34)) + (3 - -15) + -6 + -1).
+290
+Quanto é -3 - -4 - -269 - ((-16 - -5) + (22 - (8 - -3 - 8 - -14)))?
+276
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__div_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__div_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7095387ba5f607409a4739d080013254dc1159b9
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__div_big.txt
@@ -0,0 +1,20 @@
+Divide 34754707062 por -3.
+-11584902354
+Divide -1340813140 por -6.
+670406570/3
+4734892572 dividido por -4
+-1183723143
+-12017267 dividido por 41
+-12017267/41
+Calcula 29187636984 dividido por -3.
+-9729212328
+1987921 dividido por 87284
+1987921/87284
+Quanto é -31551042840 dividido por -788776071?
+40
+Divide -29643821060 por 595.
+-49821548
+Calcula 21 dividido por 1598856574.
+3/228408082
+Calcula 1133 dividido por 81117873.
+1133/81117873
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mixed_longer.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mixed_longer.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8a3a472ad7a005deb9a48e2c74858838c1443f7d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mixed_longer.txt
@@ -0,0 +1,20 @@
+Avalia (3 + 7 + (-6)/1 - (3 + 4 + -4))*((-2)/(-5))/(105/84525).
+322
+Avalia ((-16)/4)/((-32)/56) + -165 - ((-11 - -4)/(-1) - 56/8).
+-158
+Qual é o valor de (105/(-315))/((-23)/32904)*(((-600)/9)/(-10) - 6) + 4/46?
+318
+11/(-2) - (-90)/(-24420)*-1507 - (1 + (-23306)/74) - 40/10
+310
+Quanto é ((-4 - (-63)/15)/(828/(-855) - -1) - 5) + (1449472/(-6258) - (-4)/14)?
+-230
+Qual é o valor de 3 + (414/27 + 8)*(1 + (-8)/(-16)) + (1 - 3 - (2 - 2))?
+36
+Avalia 72*1*((-14)/(-8) + -4) + (-28 - (-5 - -1))*((-108)/(-864))/((-2)/(-4)).
+-168
+(12/(-56)*7*(-8)/(-6))/((176/81400)/((-2)/(-10))*(-65)/(-117))
+-333
+Avalia (-86 - ((2/(-9) - 2) + 6/27)) + (38 - -17 - 39) + -12.
+-80
+Qual é o valor de ((42/(-10))/((-4)/(-10) + (9 - 9)))/(((-87)/330 - 24/(-60))/((-2532)/4642))?
+42
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..600e363e820ead7dc0cdf808639b5e7c2c239e1a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_big.txt
@@ -0,0 +1,20 @@
+Qual é o produto de -0.06188 e -340289?
+21057.08332
+Multiplica -0.426 e -102234198.
+43551768.348
+Quanto é -3.0124 vezes 1258.96?
+-3792.491104
+Calcula -8 * 25.0413541.
+-200.3308328
+8.245289 * -51.1
+-421.3342679
+-5286 * 2152074
+-11375863164
+Multiplica -16084916214 e 0.5.
+-8042458107
+Determina -418254101 * -3.
+1254762303
+Determina -298556610 * -4.3.
+1283793423
+-0.313*40144699
+-12565290.787
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_div_multiple_longer.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_div_multiple_longer.txt
new file mode 100644
index 0000000000000000000000000000000000000000..30a030dd20ee4833cabb8cd8dcbab9389fedf86d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/arithmetic__mul_div_multiple_longer.txt
@@ -0,0 +1,20 @@
+117561/(-1052)*((-3)/(-12))/((30/(-4))/(-15))*(-16)/6*1
+149
+((130/(-39))/(4/(-54)))/(((834/1946*(-28)/(-18))/((-4)/3))/((-10)/(-4)))
+-225
+(65/(-26)*(7/(140/512)*20)/((-250)/(-10)))/(4/(-180)*18)
+128
+Avalia 325/(((-9)/(117/(-13)))/((-1)/(-1)))*2/(-5)*-2*7/49*7.
+260
+Qual é o valor de (2/(-4)*-3*(-1260)/(-2835))/((4/159)/(2*11*(-4)/(-44)))?
+53
+Qual é o valor de ((((-6)/10)/((135/(-1390))/(-9)))/(9/15))/((-62944)/(-20232)*(-6)/(-28))?
+-139
+Qual é o valor de (((1/75)/(32/192))/(((-68)/1560)/(-17)))/((-369)/(-2460))*1?
+208
+Calcula 3/(-4)*-2*(-2)/5*1*10/(-5)*1*28*(-125)/(-600).
+7
+Qual é o valor de (9*10/(-30))/(1/(-90)*6*(-4)/4*4/(-4)*(-19)/38)?
+-90
+Quanto é ((((-5)/(30/(-8)))/(-3*(-20)/90))/(97/5723))/((-42)/(-21))?
+59
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__closest_more.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__closest_more.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9af2586af6a559134d338ef3867a76f7eb827b0b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__closest_more.txt
@@ -0,0 +1,20 @@
+Qual está mais perto de -5?  (a) 5  (b) -190  (c) -1/2  (d) -1/3  (e) 3  (f) -1  (g) 15  (h) -3/5
+f
+Qual está mais próximo de -1/2?  (a) -3  (b) -0.4  (c) -362  (d) -1907  (e) -4  (f) 0.2  (g) 4
+b
+Qual valor está mais próximo de 2/3 em -0.8, -1, -91, -14, 1652, -1/8, 2/5, 5?
+2/5
+Qual valor está mais perto de 0.1 em -3, -3/2, -1466, 5, -5.2, -2/17, 598?
+-2/17
+Qual está mais perto de 1?  (a) -26/7  (b) 98  (c) -0.1  (d) -5  (e) -4/77  (f) -80.1  (g) -3
+e
+Qual valor está mais perto de -0.1 em -0.1, 0.1473, 3/7, -0.5, -39/2, 4, 3?
+-0.1
+Qual valor está mais próximo de 1/3 em -4, -0.5, 4/7, 0.41, -0.1765, 0.05, 5?
+0.41
+Qual está mais próximo de -1?  (a) 2  (b) 4  (c) -1.14  (d) -22/9  (e) -0.4  (f) -24  (g) -1.23
+c
+Qual valor está mais perto de -4 em -17/5, -25, 3/37, 3/4, 4, -13, -5?
+-17/5
+Qual está mais próximo de 1/5?  (a) -33  (b) 6/7  (c) -0.4  (d) -1.2  (e) -0.24  (f) -32  (g) 1/6  (h) 0.5
+g
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__kth_biggest_more.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__kth_biggest_more.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4e849734c2ce54f4805bbee45003f8fff6efa87c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__kth_biggest_more.txt
@@ -0,0 +1,20 @@
+Qual valor é terceiro maior?  (a) 3/7  (b) 15/142361  (c) 0.2  (d) 0  (e) -0.3  (f) 0.3  (g) 2  (h) -0.5  (i) -14
+f
+Qual valor é nono menor?  (a) 6.2  (b) -5  (c) 7  (d) 0.1  (e) -1/9  (f) -1  (g) -66  (h) 0.07  (i) -2
+c
+Qual valor é nono maior?  (a) -4  (b) -5  (c) -3  (d) 4/55  (e) 1/5  (f) -29  (g) -1  (h) -0.4  (i) 2/393
+f
+Qual valor é décimo maior?  (a) -0.1  (b) -0.2  (c) 0.3  (d) -5  (e) -3/4  (f) 0.09  (g) -55/3  (h) -0.082  (i) 1/2  (j) 3
+g
+Qual é o sexto maior valor em -5, 35, -21, -0.07, 63, 0.5, 1/7, 53, 2/9?
+1/7
+Qual é o sétimo menor valor em -3.1, -1/7, 4/3, -0.1, 2/3, -5, -9, -2, 2/15, 4/7?
+2/15
+Qual é o sétimo maior valor em -10, 0.4, 1/3, 261, -59.86, -0.1, -2/3, 2/15, 12?
+-2/3
+Qual valor é sexto menor?  (a) -2203  (b) 71  (c) 1/2  (d) -4  (e) -3/7  (f) 3  (g) 11  (h) 0.2  (i) -2
+c
+Qual é o terceiro maior valor em -5, 5, 2, -9, 51, 3/4, -3/4, -2/7, -3.2?
+2
+Qual é o segundo menor valor em 18, 0.5, 8, 33, 0.3, 9, 1, -4, 2, 26?
+0.3
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__sort_more.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__sort_more.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a9f70ab0ca9f7365613c630d543c412303ed24a7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/comparison__sort_more.txt
@@ -0,0 +1,20 @@
+Ordena 2, 11, 29, 8, 3, 450, -4, 0, 4 por ordem decrescente.
+450, 29, 11, 8, 4, 3, 2, 0, -4
+Coloca 27, 7, 5, 13, 1, 3, -1, 57, 2 em ordem decrescente.
+57, 27, 13, 7, 5, 3, 2, 1, -1
+Ordena 8, 2, 5, -8, 26, -4, 0, 4, 1 por ordem ascendente.
+-8, -4, 0, 1, 2, 4, 5, 8, 26
+Ordena 46, 0, 4, 3, 23, 6, -1, -7, -3 por ordem decrescente.
+46, 23, 6, 4, 3, 0, -1, -3, -7
+Ordena 3, -1, 57, -3, 16, 9, 11, 4, 7 por ordem ascendente.
+-3, -1, 3, 4, 7, 9, 11, 16, 57
+Ordena 2, -56, 24, 1, 13, 4, 0, 3, -3.
+-56, -3, 0, 1, 2, 3, 4, 13, 24
+Ordena -432, 5, -1, 3, 8, 17, 0, 9, 2 por ordem descendente.
+17, 9, 8, 5, 3, 2, 0, -1, -432
+Coloca 3, -1, 9, 1, 5, -2, -6, 4, -12 em ordem crescente.
+-12, -6, -2, -1, 1, 3, 4, 5, 9
+Ordena -4, 7, -1, 1, 2, 4, 857, 6, 12 por ordem decrescente.
+857, 12, 7, 6, 4, 2, 1, -1, -4
+Coloca 2, -4, 5, -18, 3, -57, 4, 8, 7 em ordem decrescente.
+8, 7, 5, 4, 3, 2, -4, -18, -57
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/measurement__conversion.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/measurement__conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5afca82f74e3c476963c4932a423eea46078e233
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/measurement__conversion.txt
@@ -0,0 +1,20 @@
+Quantos(as) minutos estão em 209891875.2 nanosegundos?
+0.00349819792
+Quanto é 24278998.5nm em quilómetros?
+0.0000242789985
+Quanto é 0.744057522 meses em milénios?
+0.0000620047935
+Quantos(as) litros estão em 26506726.8 mililitros?
+26506.7268
+Converte 44.9700523 millímetros para metros.
+0.0449700523
+Converte 0.425743507 centímetros para micrómetros.
+4257.43507
+Converte 3.24533491 anos para milénios.
+0.00324533491
+Quanto é 2763.8671 quilogramas em nanogramas?
+2763867100000000
+Quantos(as) centímetros estão em 37.5474328 quilómetros?
+3754743.28
+Quantos(as) milénios estão em 219414.232 anos?
+219.414232
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__place_value_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__place_value_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..25ac9387a9b34d2359b5c335d345e87ed11346c5
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__place_value_big.txt
@@ -0,0 +1,20 @@
+Qual é o dígito das dezena milhões de 585104582785?
+0
+Qual é o dígito das unidades de 890994362752?
+2
+Qual é o dígito das centenas de 389726456276?
+2
+Qual é o dígito das centena milhões de 669628353376?
+6
+Qual é o dígito das centenas de 847868848670?
+6
+Qual é o dígito das dezenas de 906525410143?
+4
+Qual é o dígito das dezena mil milhões de 116778722434?
+1
+Qual é o dígito das dezena milhões de 816535057076?
+3
+Qual é o dígito das dezena milhões de 294769585676?
+6
+Qual é o dígito das dezenas de 429731328864?
+6
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__round_number_big.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__round_number_big.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8611e00e9ed971341717b4e2f1ce8724694a6950
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/numbers__round_number_big.txt
@@ -0,0 +1,20 @@
+Quanto é 14091589.69193 arredondado para as 1000 mais próxima?
+14092000
+Arredonda 258672564.983 para as 100000 mais próxima.
+258700000
+Quanto é 8.30586137387 arredondado para as inteiro mais próximo?
+8
+Arredonda -383.62758343 para as casa decimal das doiss.
+-383.63
+Arredonda 21306920.88228 para as dez milhar mais próxima.
+21310000
+Arredonda -5458120.1311 para as dez milhar mais próxima.
+-5460000
+Arredonda 165358640.0843 para as um centena mais próxima.
+165358600
+Arredonda 19628.6240013 para as dez mais próxima.
+19630
+Arredonda 0.00024028921527 para as casa decimal das 5s.
+0.00024
+Arredonda 0.00222836314178 para as casa decimal das quatros.
+0.0022
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_level_set_more_samples.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_level_set_more_samples.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2afa810d54056ae4219eb9b30c7b49d3be0dde3e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_level_set_more_samples.txt
@@ -0,0 +1,20 @@
+Cinco caractéres são escolhidos sem reposição de ofnfnfofffnffnnfknn. Qual é a probabilidade de(a) escolher 3 n, 1 o e 1 k?
+35/5814
+Qual é a probabilidade de(a) escolher 4 x e 1 j quando cinco caractéres são escolhidos sem reposição de {j: 7, x: 4, s: 2}?
+7/1287
+Cinco caractéres são escolhidos sem reposição de qcqqcccccccccqc. Dá a probabilidade de(a) escolher 5 c.
+2/13
+Qual é a probabilidade de(a) escolher 1 v, 2 c, 1 n e 1 s quando cinco caractéres são escolhidos sem reposição de cvsnsvc?
+4/21
+Cinco caractéres são escolhidos sem reposição de {x: 2, l: 5, s: 5}. Dá a probabilidade de(a) escolher 1 s e 4 l.
+25/792
+Cinco caractéres são escolhidos sem reposição de denqqjeqqejjq. Dá a probabilidade de(a) escolher 2 e, 1 n e 2 j.
+1/143
+Cinco caractéres são escolhidos sem reposição de fzezzz. Dá a probabilidade de(a) escolher 4 z e 1 f.
+1/6
+Cinco caractéres são escolhidos sem reposição de {a: 2, d: 2, j: 1, b: 1, u: 1, k: 3}. Qual é a probabilidade de(a) escolher 1 j, 2 d, 1 b e 1 k?
+1/84
+Cinco caractéres são escolhidos sem reposição de bfhkfgyyfffff. Dá a probabilidade de(a) escolher 2 k, 1 g, 1 y e 1 f.
+0
+Qual é a probabilidade de(a) escolher 3 x e 2 v quando cinco caractéres são escolhidos sem reposição de {v: 4, x: 8}?
+14/33
diff --git a/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_sequence_more_samples.txt b/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_sequence_more_samples.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9af8041ec2fc1fc26c5da6cd21f873f1362a0ac2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/extrapolate/probability__swr_p_sequence_more_samples.txt
@@ -0,0 +1,20 @@
+Cinco caractéres são escolhidos sem reposição de {y: 1, k: 5, d: 2}. Qual é a probabilidade de(a) sequência kkkdy?
+1/56
+Qual é a probabilidade de(a) sequência famft quando cinco caractéres são escolhidos sem reposição de maaftammtaf?
+1/1155
+Calcula da probabilidade de(a) sequência wbwrr quando cinco caractéres são escolhidos sem reposição de wwwwwwbwrwwwwr.
+1/1092
+Calcula da probabilidade de(a) sequência jxjxy quando cinco caractéres são escolhidos sem reposição de bjjjbyxbbjxj.
+1/2376
+Calcula da probabilidade de(a) sequência vqvqv quando cinco caractéres são escolhidos sem reposição de {v: 6, q: 7}.
+14/429
+Cinco caractéres são escolhidos sem reposição de kdktddktqttfddtkdrr. Dá a probabilidade de(a) sequência ftrkt.
+1/8721
+Calcula da probabilidade de(a) sequência dovoe quando cinco caractéres são escolhidos sem reposição de oiveioooozodeo.
+1/2860
+Cinco caractéres são escolhidos sem reposição de {x: 11, y: 2}. Dá a probabilidade de(a) sequência xxxyx.
+4/39
+Cinco caractéres são escolhidos sem reposição de {q: 2, n: 8, u: 3, h: 6}. Qual é a probabilidade de(a) sequência nnunu?
+7/4845
+Cinco caractéres são escolhidos sem reposição de {q: 6, i: 4, b: 7}. Dá a probabilidade de(a) sequência iiqbi.
+3/2210
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9bc185ff571f57ba838912fd4d82901be43b0839
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d.txt
@@ -0,0 +1,20 @@
+Resolve -31*b = 42*b - 2 + 3360 para b.
+-46
+Resolve 0 = 79*w - 174 - 225 - 75 para w.
+6
+Resolve 0 = -2363*b + 2158*b - 1845 para b.
+-9
+Resolve 27271 - 11807 + 29840 = 1618*o para o.
+28
+Resolve -178*x - 14124 = -104*x - 502*x para x.
+33
+Resolve 2069 - 965 = -109*r - 145*r - 11088 para r.
+-48
+Resolve 1168*j - 14076 + 5317 = -3311 + 14408 para j.
+17
+Resolve 201*n + 432*n - 11137 = 148*n + 13113 para n.
+50
+Resolve 13405*q - 13462*q = 1710 para q.
+-30
+Resolve 0 = -4827*o + 3861*o + 2898 para o.
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3bb8732b77036262738fa6aa4c8a5429b0c89370
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_1d_composed.txt
@@ -0,0 +1,20 @@
+Seja k = 901 - 894. Supõe v - 4 - 4 = 0. Resolve k*b = v + 6 para b.
+2
+Assume que z é ((-483)/(-1288))/(6/112). Resolve -163 = z*u - 177 para u.
+2
+Supõe -11 = -2*z + 29. Assume que t é 14/4 - (z/8 - 1). Resolve -8 = 2*o + t para o.
+-5
+Supõe -2*y - 12 = -i - 13, -4*y + 11 = i. Seja z(m) = m**3 + 3*m**2 - 4*m. Seja o, z(-4). Supõe 3*j - 69 = -o*j. Resolve -4*g = i - j para g.
+5
+Assume que a é 2474*(-1)/(-26) - (1236/(-78) - -16). Resolve 3*g + 16*g + a = 0 para g.
+-5
+Supõe 13*f + 13*f + 287 = 33*f. Resolve 46 = -29*a - f para a.
+-3
+Seja f = 1 - -17. Seja l(h) = -h**3 + 8*h**2 - 9*h - 12. Seja s(t) = t + 1. Seja d(b) = -l(b) - 3*s(b). Seja x, d(7). Resolve -4*i - f = -x para i.
+-4
+Supõe -1 = 3*r + 3*s + 8, -3*r + 6 = -2*s. Seja g(f) = 2*f**2 + 7*f + 3. Seja n, g(-5). Assume que y é (r - 2)/((-9)/n). Resolve y*c - 7 - 9 = 0 para c.
+4
+Seja t = 11 + -17. Assume que l é t/(-6) + (-4)/(-1). Seja f(u) = 2*u**2 - 2*u - 1. Seja q, f(-1). Resolve -q = 4*r + l para r.
+-2
+Seja u = -43 + 33. Seja s = u + 11. Assume que v é ((-3)/(-2) - s)/((-39)/(-702)). Resolve 5*w + 1 = -v para w.
+-2
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..74ad6f6f5c8efc7d0d928043d07d92c6d6fbec87
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d.txt
@@ -0,0 +1,20 @@
+Resolve -h - 1 = -9526*b + 9527*b, -b = -10*h - 10 para h.
+-1
+Resolve 4*x = -14*z - 590, 88 = -2*z + 2 para x.
+3
+Resolve 0 = -6*c - 9*o + 12*o - 3, 32*o + 167 = -4*c + o para c.
+-3
+Resolve -5*s = 5*v + 135, -2*s + s - 58*v + 55*v = 27 para s.
+-27
+Resolve -1539*c + 1543*c - 60 = -4*w, -108*c = -4*w - 388 para w.
+11
+Resolve -3*v = 3*p + 631 - 619, 23 = -5*p - 4*v - 0*v para p.
+-7
+Resolve 0 = 50*u + 2*v - 698 - 44, 3*u = 5*v + 65 para u.
+15
+Resolve 3*j = -4*s + 11, 0 = j + 1957 - 1962 para s.
+-1
+Resolve -4*d - 3798*t + 52 = -3776*t, -3*d - 4*t + 14 = 0 para d.
+2
+Resolve -4*l + 7*c = -392 + 416, -3*c - 19 = -31*l para l.
+1
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..bc0c668f98869f30257df51fb6eff1fc6e83fa92
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__linear_2d_composed.txt
@@ -0,0 +1,20 @@
+Supõe -a - 5*j + 99 = 0, 2*a + 5*j - 56 = 167. Seja d = -115 + a. Resolve -2*v - d = -5*v, 3*i - v = 0 para i.
+1
+Supõe -10 = -j + 6. Supõe -3*o + 8*o + j = h, 5*o = -3*h + 8. Seja s = -4 + h. Resolve -d + 4*f + 16 = 0, -s*f - 6 = -0 para d.
+4
+Assume que w é 3/15 - 0 - (-2247)/15. Assume que a é 127/6 + (-25)/w. Resolve -3*z + a = 2*d, 0*d - 5*z = 3*d - 34 para d.
+3
+Seja r = -2599 - -2604. Resolve -10 = 5*t, 18 = -r*z + 8*t - 12*t para z.
+-2
+Supõe 64*b + 60 - 368 = -13*b. Resolve -b*r - 4*u = 20, -5*r - 2*u + 9 = 31 para r.
+-4
+Seja w(a) = -a**3 + 5*a**2 + 27*a - 17. Seja b, w(8). Resolve -g - b = 12*y - 9*y, y - 11 = 3*g para y.
+-1
+Seja w(h) = -49*h**3 + 2*h**2 - h - 1. Seja t, w(1). Assume que j é 4/14 - (-760)/(-14). Seja q = t - j. Resolve -4*m + z + q = -8, 16 = 5*m - z para m.
+3
+Supõe -15*l + 6*l - 3 = -3. Resolve 16 = u + 3*p, l = -4*u + 3*p + 4 para u.
+4
+Seja i(f) = 63 - f. Seja m, i(16). Seja z = m - 34. Resolve 5*u + 0*r = -3*r - 5, -u - z = 3*r para u.
+2
+Seja n(g) = 1390 - 33*g. Seja l, n(42). Resolve 12*f + l*d + 3 = 11*f, -3*f = 3*d para f.
+1
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c4733009cbef057675e30f3ab1a12adde3372455
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots.txt
@@ -0,0 +1,20 @@
+Resolve 243*f**5/7 + 314823*f**4 + 4988715660*f**3/7 - 35139394944*f**2/7 + 30889839264*f/7 - 741363984/7 = 0 em relação a f.
+-4538, 2/81, 1, 6
+Factoriza -n**3 - 15936*n**2 - 984125*n + 1000062.
+-(n - 1)*(n + 63)*(n + 15874)
+Factoriza -5*g**2 + 13885*g - 8809380.
+-5*(g - 1796)*(g - 981)
+Resolve -2*t**4/9 - 1970*t**3/9 + 9982*t**2/9 + 9010*t - 9900 = 0 em relação a t.
+-990, -5, 1, 9
+Seja 2436*g**2 - 187564*g - 616 = 0. Calcula g.
+-2/609, 77
+Factoriza 5*a**2 + 52935*a - 2819600.
+5*(a - 53)*(a + 10640)
+Encontra d tal que -4*d**2/7 + 6104*d - 340032 = 0.
+56, 10626
+Seja 11*o**3 - 5661*o**2 + 28030*o = 0. Calcula o.
+0, 5, 5606/11
+Supõe -4*k**3/7 - 37752*k**2/7 - 37740*k/7 + 75496/7 = 0. Qual é k?
+-9437, -2, 1
+Supõe 4*y**3 - 507180*y**2 + 16075704000*y + 80391200000 = 0. Calcula y?
+-5, 63400
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fd701d2a178991a5da7c43d3d5e5bb37aa35c353
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__polynomial_roots_composed.txt
@@ -0,0 +1,20 @@
+Seja l(q) = 30*q + 336. Seja a, l(-11). Seja u(v) a segunda derivada de 0*v**3 + 0 + 1/3*v**a + 0*v**2 + 5*v - 7/10*v**5 + 1/3*v**4. Factoriza u(c).
+2*c**2*(c - 1)*(5*c - 2)
+Supõe -43*i = -65*i. Seja s(q) a terceira derivada de -1/140*q**5 + 0*q - 25/14*q**3 + 18*q**2 + i + 5/28*q**4. Resolve s(d) = 0.
+5
+Seja y(z) a terceira derivada de -2*z**7/105 + 181*z**6/60 + 139*z**5/15 - 46*z**4/3 - 70*z**2 - 2*z. Seja y(t) = 0. Calcula t.
+-2, 0, 1/2, 92
+Seja x(f) a segunda derivada de -f**4/12 + 13*f**3/36 + 5*f**2/4 + 1050*f. Factoriza x(a).
+-(a - 3)*(6*a + 5)/6
+Seja y(q) = -153*q**3 - 255*q**2 - 222*q - 25. Seja w(p) = 31*p**3 + 51*p**2 + 45*p + 5. Seja f(r) = 19*w(r) + 4*y(r). Qual é o valor de c em f(c) = 0?
+-1, -5/23
+Assume que a é ((-20)/(-3))/((-25)/(-30)). Assume que c é ((2/a)/1)/((-1110)/(-4736)). Determina y, dado que c*y + 0 - 4/15*y**3 - 2/15*y**4 + 8/15*y**2 = 0.
+-2, 0, 2
+Seja v(s) a terceira derivada de s**6/60 + 13*s**5/18 + 149*s**4/36 - 19*s**3/3 - 1637*s**2. Factoriza v(z).
+2*(z + 3)*(z + 19)*(3*z - 1)/3
+Seja o(m) a primeira derivada de 1/2*m**2 + 0*m**3 + 15*m - 1/12*m**4 + 8. Seja h(f) a primeira derivada de o(f). Encontra x, dado que h(x) = 0.
+-1, 1
+Seja s(a) a segunda derivada de a**5/30 + 25*a**4/9 + 97*a**3/9 + 16*a**2 + 2*a - 344. Factoriza s(r).
+2*(r + 1)**2*(r + 48)/3
+Assume que b é (-8 + 2)*(-1)/(-3) + 40/18. Supõe 8 = -w + 5*w. Factoriza b*r**w - 8/9 + 0*r.
+2*(r - 2)*(r + 2)/9
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_next_term.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_next_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3fae20bd7a7da61636b8caf89244479c422fa668
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_next_term.txt
@@ -0,0 +1,20 @@
+Qual é o termo seguinte em -1506717, -1506716, -1506713, -1506708, -1506701?
+-1506692
+Qual é o termo seguinte em 361606, 723207, 1084808, 1446409, 1808010?
+2169611
+Qual é o termo seguinte em 330, 2680, 9066, 21504, 42010, 72600?
+115290
+Qual é o termo seguinte em 25159, 50326, 75463, 100564, 125623, 150634, 175591?
+200488
+O que vem a seguir: 3061, 1652, 243?
+-1166
+Qual é o termo seguinte em 251, 542, 903, 1334?
+1835
+Qual é o termo seguinte em -60084, -60079, -60074, -60069, -60064?
+-60059
+Qual é o termo seguinte em -4282026, -8564064, -12846110, -17128170, -21410250, -25692356, -29974494?
+-34256670
+O que se segue em 4900, 9789, 14636, 19441?
+24204
+O que se segue em -514769, -1029532, -1544293, -2059052, -2573809, -3088564, -3603317?
+-4118068
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_nth_term.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_nth_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f3240b05e976165aa30cee5bd207e077fe2a65ce
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/algebra__sequence_nth_term.txt
@@ -0,0 +1,20 @@
+Qual é termo nº h na sequência -1330810, -2661626, -3992442?
+6 - 1330816*h
+Qual é termo nº n na sequência -341601, -341586, -341563, -341532, -341493?
+4*n**2 + 3*n - 341608
+Qual é termo nº y na sequência 1022123, 4088490, 9199103, 16353962, 25553067, 36796418?
+1022123*y**2 - 2*y + 2
+Qual é termo nº a na sequência -1570, -3187, -4786, -6361, -7906?
+a**3 + 3*a**2 - 1633*a + 59
+Qual é termo nº y na sequência 20043, 80589, 181491, 322743, 504339, 726273, 988539?
+-y**3 + 20184*y**2 + y - 141
+Qual é termo nº q na sequência -3820, -6819, -9832, -12865, -15924, -19015?
+-q**3 - q**2 - 2989*q - 829
+Qual é termo nº l na sequência 253463, 506920, 760373, 1013822, 1267267, 1520708?
+-2*l**2 + 253463*l + 2
+Qual é termo nº d na sequência -9077440, -18154880, -27232320, -36309760?
+-9077440*d
+Qual é termo nº x na sequência 20953, 83859, 188699, 335467, 524157, 754763?
+-x**3 + 20973*x**2 - 6*x - 13
+Qual é termo nº z na sequência -742187, -742231, -742305, -742409, -742543?
+-15*z**2 + z - 742173
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub.txt
new file mode 100644
index 0000000000000000000000000000000000000000..06b5d6336043d963b8ce582b733e03e5637aa069
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub.txt
@@ -0,0 +1,20 @@
+Determina 19283.956636 + -0.05.
+19283.906636
+Determina 63.7789 - -20052.
+20115.7789
+Determina -0.04 + -37488652513.
+-37488652513.04
+Calcula 71143946 - 12.31.
+71143933.69
+Quanto é 49430807153 menos 0.2?
+49430807152.8
+Quanto é -53.497735 - -11.6?
+-41.897735
+Determina 0.126321789662 + -0.4.
+-0.273678210338
+O total de -0.1 mais -1237.9494883.
+-1238.0494883
+Determina -0.4 - 15293176390.
+-15293176390.4
+Quanto é -718.82 menos -843323?
+842604.18
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub_in_base.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub_in_base.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2f96b9d721f969e1cb71283bd6074d456e4927ca
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_or_sub_in_base.txt
@@ -0,0 +1,20 @@
+Em base 8, quanto é 463 + -46277?
+-45614
+Em base 15, quanto é 19da490 + 5?
+19da495
+Em base 9, quanto é 83241 + -56?
+83174
+Em base 8, quanto é 11576 + 454?
+12252
+Em base 13, quanto é 59 + 5587c?
+55908
+Em base 5, quanto é 1 - 2211334240?
+-2211334234
+Em base 12, quanto é -4 - -168b5a6?
+168b5a2
+Em base 9, quanto é 55683680 + -2?
+55683677
+Em base 14, quanto é -c + -194955?
+-194963
+Em base 3, quanto é 2121 - 11010011?
+-11000120
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_sub_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_sub_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..afb6fa6520228a025ca4b3e1ebd0378cdb92481f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__add_sub_multiple.txt
@@ -0,0 +1,20 @@
+Quanto é -21 - (-7 + 1 + (40 - 104))?
+49
+5 + 0 - (16 - 28 - -27 - (-11 + 22))
+1
+Calcula -17 + (0 - (19 - 72)).
+36
+-15 + (-8 - -15) + 6 + -23 + 1 + 1
+-23
+Quanto é 11 + (12 - 1 - -1 - (5 - 1))?
+19
+Avalia 40 + (-2 + 1 - 5) + (0 - -6) + 5.
+45
+Calcula (1 - 29 - (-130 - -137)) + -4.
+-39
+Quanto é 3 - (1 + 34 + (-35 - -21))?
+-18
+(0 - -9) + (-4 - (-40 - (-15 - 2)))
+28
+Avalia (-119 - (44 + -72)) + 42.
+-49
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__div.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__div.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1c1436be477a5d97b95fbbf22b400b23f2e834c6
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__div.txt
@@ -0,0 +1,20 @@
+Calcula 298 dividido por -12183.
+-298/12183
+Quanto é 9739216 dividido por 4869608?
+2
+8600496 dividido por -2866832
+-3
+Divide 0 por -2900681.
+0
+875120 dividido por -16
+-54695
+-7010715 dividido por 4845
+-1447
+Quanto é 15299204 dividido por 24716?
+619
+Divide 0 por -3245315.
+0
+35861 dividido por -444
+-35861/444
+Calcula 115125165 dividido por 5.
+23025033
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mixed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mixed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..856b8a0108b4f2b936d8124257c2178907bdc442
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mixed.txt
@@ -0,0 +1,20 @@
+Calcula (((-5)/6 + (-114)/684)*-1)/((-201)/(-6)).
+2/67
+Calcula (75/10 - 10) + 668/(-56) - (-8)/(-14).
+-15
+1736/(-98) + 18 - ((-8)/4 - 342/(-21))
+-14
+Quanto é (32*(-22 + 10489/476))/((-2)/(-7))?
+4
+101 + -97 + -11 + -38
+-45
+Avalia 1158/32 - (1 + 39/(-48)).
+36
+-3 + 1202/208 + 3/78*(-16 + 12)
+21/8
+Qual é o valor de ((-14)/8)/(-5 - (202/25)/(16/(-10)))?
+-35
+Quanto é (-7496)/(-243620) - (2 - 583/65)?
+7
+Quanto é (-3)/((-39)/(-2)) + ((-187368)/1950 - -96)?
+-6/25
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fca42a5011a7d811284df22af42e2a46483b5388
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul.txt
@@ -0,0 +1,20 @@
+15836.2*-0.48
+-7601.376
+-0.18 vezes 10345.65
+-1862.217
+Produto de -541134 e 1.5.
+-811701
+Produto de -9.1 e 0.111539.
+-1.0150049
+-0.5*-0.3542049
+0.17710245
+Calcula -1.79239 * -0.11.
+0.1971629
+Multiplica -0.1 e -237.9431.
+23.79431
+71418*-17
+-1214106
+Produto de 0.4 e -3459159.
+-1383663.6
+Multiplica -69782 e 176.
+-12281632
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul_div_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul_div_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..35938891756fd6a5a60f0d6742c668bd31273ded
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__mul_div_multiple.txt
@@ -0,0 +1,20 @@
+406/(-377)*(-5187)/(-114)
+-49
+Calcula (108/14)/((6/(-168))/(136/(-1088))).
+27
+Quanto é (-20*(-258)/(-4300))/(6/40*1/1)?
+-8
+Quanto é ((-1296)/(-378))/(2/(-28))?
+-48
+Quanto é (-6)/((-576)/60)*(48/(-8))/(-3)?
+5/4
+Quanto é (-3)/(-6)*(60/700*-10)/((-3)/189)?
+27
+Qual é o valor de ((-9084)/39364)/((-6)/(-104))?
+-4
+Quanto é (-15)/(-21)*48/(-40)*-1*4/16?
+3/14
+-44*(-56)/(-53900)*35
+-8/5
+Qual é o valor de (-50)/(-825)*-11*(-2)/120*-4?
+-2/45
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__nearest_integer_root.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__nearest_integer_root.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e9e3c3e70da9299425db7d7bd42b5b16615a4cbc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__nearest_integer_root.txt
@@ -0,0 +1,20 @@
+Quanto é 79421780 elevado a 1/2, arredondado para o inteiro mais próximo?
+8912
+Quanto é 50308105 elevado a 1/2, arredondado para o inteiro mais próximo?
+7093
+Qual é a raíz cúbica de 65957304 arredondada para o inteiro mais próximo?
+404
+Quanto é 47181849 elevado a 1/2, arredondado para o inteiro mais próximo?
+6869
+Qual é a décima raíz de 5923664 arredondada para o inteiro mais próximo?
+5
+Qual é a terceira raíz de 42316370 arredondada para o inteiro mais próximo?
+348
+Qual é a sexta raíz de 70908946 arredondada para o inteiro mais próximo?
+20
+Quanto é 55976877 elevado a 1/9, arredondado para o inteiro mais próximo?
+7
+Quanto é 41901349 elevado a 1/8, arredondado para o inteiro mais próximo?
+9
+Qual é a raíz cúbica de 72976644 arredondada para o inteiro mais próximo?
+418
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__simplify_surd.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__simplify_surd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5ffae6b52e0f5a3ea90e103ec22a464e9700d2fe
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/arithmetic__simplify_surd.txt
@@ -0,0 +1,20 @@
+Simplificar ((1*-5*(-1*(2*sqrt(891) + sqrt(891)) + 5 + sqrt(891)*-1) + 1)*3)**2.
+3212784 - 77760*sqrt(11)
+Simplificar (-3*(sqrt(270) + sqrt(270)*-1*4))/((sqrt(540)/sqrt(2))/(sqrt(12)/sqrt(4)*6)).
+54*sqrt(3)
+Simplificar -5*-1*(-3 + (-5 + (-1 + sqrt(704) - sqrt(704)))**2)*-5 + 2.
+-823
+Simplificar (((0 + sqrt(125))*-2)**2*6 - (sqrt(180)*2)**2*-5)*-4.
+-26400
+Simplificar (-5*(5 + sqrt(275)*1 + -5))**2 + -2*sqrt(275) + 4 + (0 + sqrt(275) + 2)**2.
+10*sqrt(11) + 7158
+Simplificar -5*((sqrt(32) - (sqrt(32) + -3*sqrt(512)))**2 + -3 - (sqrt(800)*-1*4)**2).
+40975
+Simplificar 3 + ((2*sqrt(243) + sqrt(243) - sqrt(243))**2*4 - ((sqrt(243) + 1)*1)**2) - -5*(-2*(sqrt(243) + 0) - (sqrt(243) + 5*(sqrt(243) + 0))**2).
+-40093 - 108*sqrt(3)
+Simplificar 4*((-2 + sqrt(112)*-2)*6 + -4 + (sqrt(112) - (1 + sqrt(112))) + -4) + 0.
+-192*sqrt(7) - 84
+Simplificar -5 + -2 + (0 + -3 + sqrt(80) + 0 - ((sqrt(80) - (-1*sqrt(80))**2) + -5 + 0 + (sqrt(80)*-1)**2)).
+-5
+Simplificar -3 + (3*(sqrt(153)/(-2*sqrt(9)) + sqrt(17)) - (-1*sqrt(85)*-4)/sqrt(5))**2 + 2.
+421/4
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2753899a27e05ffb41364936302f5c7a4b2f4a00
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate.txt
@@ -0,0 +1,20 @@
+Encontra a terceira derivada de 2564804*a**4 - 518*a**2 + 12756*a em relação a a.
+61555296*a
+Encontra a segunda derivada de s**3 - 30529*s**2 + 6926614*s em relação a s.
+6*s - 61058
+Encontra a segunda derivada de 333833*s**2*z**2 + 3*s**2*z - 21663*s*z - 8*z**4 + 31*z + 2 em relação a z.
+667666*s**2 - 96*z**2
+Qual é a segunda derivada de 7957*a**2*v**2*y + 4*a**2*y + 4*a*v*y**3 + 719*v**2*y**3 + 4142*v*y em relação a v?
+15914*a**2*y + 1438*y**3
+Qual é a terceira derivada de 4555153*n**3 + 3*n**2 - 3387*n + 2?
+27330918
+Encontra a segunda derivada de -2819*s**3 + 241*s**2 + 2237*s + 2745 em relação a s.
+482 - 16914*s
+Encontra a terceira derivada de 27473*j**6 - 100*j**5 + 77504*j**2 - 8*j - 2.
+3296760*j**3 - 6000*j**2
+Encontra a segunda derivada de -11974623*r**2 + 811338*r.
+-23949246
+Qual é a terceira derivada de -8306400*z**3 - 339460*z**2 + 13*z?
+-49838400
+Qual é a segunda derivada de -18*b**3 - 243738*b**2 + 340*b - 2983?
+-108*b - 487476
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1478af77a5fd0c4cb72cbc69a4765f1883f17463
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/calculus__differentiate_composed.txt
@@ -0,0 +1,20 @@
+Assume que s é ((-31)/2)/(4/(-8)). Qual é a segunda derivada de 20*t**3 - 7 + 98*t - 95*t + s*t**3 - 6*t**3 em relação a t?
+270*t
+Supõe -78 = r + 2*r. Seja q = -22 - r. Encontra a terceira derivada de 0*k**4 - 6*k**2 - 4*k**q - 6*k**2 + 0*k**2 em relação a k.
+-96*k
+Qual é a segunda derivada de -312*o**4 - 354*o**4 + 18431*o**3 + 18420*o**3 + 838*o + 2*o**2 - 36851*o**3 em relação a o?
+4 - 7992*o**2
+Seja t(o) a terceira derivada de -16/3*o**3 + 0*o**5 + 5/24*o**4 - 37*o**2 - 1 + 0*o - 7/120*o**6. Diferencia t(l) em relação a l.
+5 - 21*l**2
+Seja k(o) a segunda derivada de -71/2*o**2 - 1/10*o**5 + 11/6*o**3 - 52*o + 0 + 0*o**4. Deriva k(x) em relação a x.
+11 - 6*x**2
+Assume que v é (152/(-10))/((-8)/20). Assume que h é (-72)/(-20) + 2/5. Qual é a primeira derivada de -j + 48*j**h - v*j**4 + j - 16 em relação a j?
+40*j**3
+Seja l(o) a terceira derivada de 52*o**5/15 - 119*o**4/8 - o**3/3 - 887*o**2. Qual é a segunda derivada de l(w) em relação a w?
+416
+Supõe 446*p - 444*p = 4. Encontra a primeira derivada de p*a + 326 - 2*a - 291 - 41*a**4 em relação a a.
+-164*a**3
+Seja x(c) a primeira derivada de c**7/120 + c**5/24 + 38*c**3/3 + 70. Seja n(f) a terceira derivada de x(f). Encontra a segunda derivada de n(s) em relação a s.
+42*s
+Seja b(c) a segunda derivada de -532*c**6/15 + c**4/6 - 135*c**2 - 1142*c. Encontra a terceira derivada de b(y) em relação a y.
+-25536*y
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1ab5404a3f072979dddc1077104bdf1e042dca82
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest.txt
@@ -0,0 +1,20 @@
+Qual valor está mais perto de 5/19 em 4, 2/3, 6989?
+2/3
+Qual valor está mais perto de -11/12 em 3, -3/2449, -13?
+-3/2449
+Qual valor está mais perto de -136 em 0.4, 0, -5/2, -1/245, -2/7?
+-5/2
+Qual está mais próximo de -12/47?  (a) 0.01  (b) -16  (c) 4  (d) -5/8
+a
+Qual valor está mais próximo de 1/3 em -2/23, -13120, 3?
+-2/23
+Qual valor está mais próximo de 0 em 4765, 0.29, 23, 2/3?
+0.29
+Qual está mais próximo de -16?  (a) -2  (b) -4.4  (c) 4/9  (d) -89  (e) -4
+b
+Qual está mais perto de 2/3?  (a) 19671  (b) -1.1  (c) 0.7
+c
+Qual está mais próximo de -2?  (a) 11  (b) -3.8  (c) 0.4  (d) 99  (e) 0
+b
+Qual valor está mais perto de 0 em 9, -4, 0.02, 1515, -1?
+0.02
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8e47d819affa95e0b34509ac04c6e2fd1d8cb3c1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__closest_composed.txt
@@ -0,0 +1,20 @@
+Assume que s é (9 - 10) + -8 - (-102)/12. Supõe 2*m + 1 - 7 = 0. Seja a = m + -3. Qual valor está mais perto de 0.1 em s, -3, a?
+a
+Assume que g é (-178)/(-356)*-3*(-2)/9. Qual valor está mais próximo de g em -2, -14, 1/3, 2?
+1/3
+Seja z = -1.617 + 2.617. Qual está mais próximo de 8?  (a) -1/4  (b) z  (c) 0.4
+b
+Seja y = 4228323/169330 + -7222/295. Seja g = 5/82 - y. Qual está mais perto de -0.1?  (a) 0  (b) g  (c) -3
+a
+Seja m = 48855 + -341983/7. Assume que z é 1 + (-10)/12 + 0. Qual está mais perto de 0?  (a) z  (b) -1  (c) m
+a
+Seja y = -2595 + 2600. Qual está mais perto de -1?  (a) -1/5  (b) 1/4  (c) 3/2  (d) y
+a
+Seja k = 19609 + -19610. Supõe 4*j = 4*m + 16, -4*m - 3*j = -j - 14. Qual está mais perto de 2/9?  (a) -3  (b) m  (c) k
+b
+Seja p(z) = 3*z + 3. Seja h, p(-4). Assume que m é h/(-12)*2*2. Seja f = -83/3 + 412/15. Qual está mais perto de 2?  (a) -1  (b) f  (c) m
+c
+Seja k = 1 - 1.3. Assume que x é 4/7 - ((-3750)/(-315) + -11). Qual está mais perto de k?  (a) -1/2  (b) -8/7  (c) x
+c
+Seja f = 239/7 - 34. Supõe 4*x + 40 = -2*i, -5*x = -4*i + 9 + 15. Qual está mais perto de 0?  (a) 1  (b) x  (c) f
+c
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c17819363a42b14cfa04fa3fd608116bcabe16bc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest.txt
@@ -0,0 +1,20 @@
+Qual valor é quarto maior?  (a) -318/7  (b) 98  (c) 0.4  (d) 5  (e) -0.1  (f) -3
+e
+Qual é o terceiro menor valor em 0.3, 132, 40113?
+40113
+Qual valor é sexto maior?  (a) 3/4  (b) 315  (c) 61  (d) 4/7  (e) 1/11  (f) -1  (g) 2
+e
+Qual valor é segundo menor?  (a) 23  (b) -9  (c) 957
+a
+Qual valor é terceiro menor?  (a) -0.2  (b) -4/31  (c) 0.5  (d) -1  (e) -5  (f) 13  (g) -2
+d
+Qual é o menor valor em 10/11, 11/6, 2, -8, -1?
+-8
+Qual é o segundo menor valor em 2.3, -3, -2/9, -0.2, -4/445, -4?
+-3
+Qual valor é quarto menor?  (a) 0.08  (b) -3/5  (c) -3133  (d) 4  (e) 0  (f) 0.4
+a
+Qual é o menor valor em -756, -4, 7.489?
+-756
+Qual valor é sexto menor?  (a) 3/2  (b) -1  (c) 2.1  (d) 0.4  (e) 0.15  (f) 5  (g) -0.219
+c
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1e0e2994e8847844c61e1e2a89cd479ce8d7f240
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__kth_biggest_composed.txt
@@ -0,0 +1,20 @@
+Assume que h é (4/(-24) - -1) + (-1)/(18/(-12)). Seja x = -0.3 + 3.3. Qual é o terceiro menor valor em x, 4, h, 0?
+x
+Supõe 17 - 19 = -i, -4*f + 3*i + 14 = 0. Seja d = 0.07 + -0.16. Qual valor é terceiro maior?  (a) f  (b) d  (c) -1/2
+c
+Seja k = -106.979 - -107. Seja r = k - -1.079. Qual valor é segundo maior?  (a) -1  (b) r  (c) -0.4
+c
+Supõe -n + w = -3*w - 12, 0 = 4*n - 4*w - 24. Seja b = n + -1. Assume que r é (-3 - (-9)/b)/(-2). Qual é o terceiro maior valor em -10, r, 2?
+-10
+Assume que f é 2/(3/(-6)*2). Seja v = 323410/41 - 7888. Qual valor é terceiro maior?  (a) -1  (b) f  (c) v  (d) 1
+a
+Seja m = 199 - 202. Seja n = 1 - m. Qual é o quarto menor valor em 5/3, 0, 2, n?
+n
+Seja f = -1651 + 1648.842. Seja h = f + 0.158. Seja j = 0 + -0.1. Qual é o terceiro maior valor em h, 10, j?
+h
+Seja m = 0 + 0. Seja q = 10.87 + -155.96. Seja t = 145 + q. Qual é o menor valor em t, -1, -2, m?
+-2
+Seja s = 2228.1 + -2228. Qual é o segundo menor valor em s, 0, -83.2?
+0
+Supõe -2*f = -2*w + 2, -4*f + 0 = 2*w - 32. Supõe 0 = -w*s - 0. Supõe n + 1 - 2 = s. Qual valor é menor?  (a) -3  (b) 4  (c) n
+a
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5ed51a34dd0553a6f985fd81628ca44cad9e5267
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair.txt
@@ -0,0 +1,20 @@
+Será que 0.1 e 14906487 possuem valores diferentes?
+True
+Será 23715, no máximo, 23703?
+False
+Qual é menor: -229647 ou -229623?
+-229647
+Qual é maior: -0.3 ou 8664526?
+8664526
+Qual é menor: -1 ou -458/29429?
+-1
+Qual é menor: 728 ou -3436?
+-3436
+Será que 277047 != 277079?
+True
+Qual é maior: -114258 ou -114299?
+-114258
+Será 4351/53 menor que 81?
+False
+Será que 988 e 206318/209 não são equivalentes?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d56f9f8015e4e347e98d6714295f45ff1b467145
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__pair_composed.txt
@@ -0,0 +1,20 @@
+Assume que i é (199/(-2) + -1)/((-33)/88). Será que 268 e i não são equivalentes?
+False
+Assume que z é (-196)/(-12) + (-2)/6. Seja h = z + -16. Supõe h = 4*r - 2 - 2. Qual é maior: r ou 7/4?
+7/4
+Assume que b é (3/(-6))/((-2)/(-12)). Assume que k é (-177)/(-585) - b/(-9). Supõe 5*y + 10 = -15*h + 16*h, -5*y - 15 = -2*h. Será y <= k?
+True
+Seja h = 3120541437374/5565 - 560744221. Seja m = h + 2/1113. Qual é superior: -28 ou m?
+m
+Seja i(v) = -3*v**2 - 100*v - 115. Seja c, i(-32). Qual é menor: 96/7 ou c?
+c
+Seja h = -4842 + 4801. Será que 8 = h?
+False
+Supõe -3*p + 2*p + 2*m = -25, m + 134 = 5*p. Seja c(f) = 9 + 20*f - 21*f + 37. Seja g, c(19). Será g maior do que p?
+False
+Supõe l = 7*h - 2*h + 27, -4*l - 3*h - 7 = 0. Assume que a é 2*1646/6600 - l/4. Seja k = -10729/3300 - a. Qual é superior: -3 ou k?
+-3
+Seja u(k) = k**2 + k - 4. Assume que q é 0/(1/((-4)/8)). Seja t(g) = 2*g**3 + g**2 - g - 5. Seja b, t(q). Seja r, u(b). Qual é superior: r ou 17?
+17
+Seja r = 5142 - 5144. Supõe -2*j + 3*w = 5*w + 18, j = -2*w - 6. Será j >= r?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7ee27d42b3a36f4228e2f29e257c82e080714c81
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort.txt
@@ -0,0 +1,20 @@
+Coloca -350, -57, -3, 2 em ordem ascendente.
+-350, -57, -3, 2
+Coloca -2, 4, -0.3, 0, -1078, 1/4 em ordem ascendente.
+-1078, -2, -0.3, 0, 1/4, 4
+Coloca 2/75, -1, 1188, 2/19 em ordem decrescente.
+1188, 2/19, 2/75, -1
+Ordena 1, -3, -4, 3062, -1, -2, -5.
+-5, -4, -3, -2, -1, 1, 3062
+Ordena -11, -2, 1020, 3, 4, 8 por ordem decrescente.
+1020, 8, 4, 3, -2, -11
+Ordena -4, -0.05, 4, -0.3, -3, -783 por ordem decrescente.
+4, -0.05, -0.3, -3, -4, -783
+Ordena -22, -1, 10890, 3, -2 por ordem crescente.
+-22, -2, -1, 3, 10890
+Ordena 10/11, -0.1806, 1, -0.05 por ordem ascendente.
+-0.1806, -0.05, 10/11, 1
+Coloca -4, 13, 5, -5, 2, 3, -42 em ordem descendente.
+13, 5, 3, 2, -4, -5, -42
+Ordena -23, -1/9, -4, -5, -15, -6.
+-23, -15, -6, -5, -4, -1/9
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..546937c8da53f5c7dbd4d9585a7012230030fa0b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/comparison__sort_composed.txt
@@ -0,0 +1,20 @@
+Seja l = -6.7 + 19.2. Seja u = -17.5 + l. Ordena 3, -5/4, 0.3, u por ordem crescente.
+u, -5/4, 0.3, 3
+Supõe -25*z = -11 + 11. Coloca z, 5, -536 em ordem decrescente.
+5, z, -536
+Supõe 95 = -42*u + 179. Seja p(y) = y**2 + 3*y - 1. Seja t, p(-4). Ordena -5, t, u, 4.
+-5, u, t, 4
+Seja q = 122/27 - 158/27. Coloca -68, q, -1 em ordem crescente.
+-68, q, -1
+Seja c = 23756.5 - 23756. Seja k = -0.1 - -0.1. Seja b = k + -0.3. Coloca b, 0.08, c em ordem decrescente.
+c, 0.08, b
+Seja g = 11316/5 + -2263. Coloca g, 0.1, -0.2, -0.18 em ordem crescente.
+-0.2, -0.18, 0.1, g
+Supõe -2*n + 12*o - 8*o = 6, 0 = 4*n - 2*o - 12. Coloca n, -54, 0, 2, -3 em ordem crescente.
+-54, -3, 0, 2, n
+Seja b = 621 + -564. Ordena 1, b, 2 por ordem descendente.
+b, 2, 1
+Seja l = -3813 - -3812. Coloca 1, l, -1/11 em ordem decrescente.
+1, -1/11, l
+Supõe -32*k + 22 = -34 - 40. Ordena -5, 5, 4, -24, k.
+-24, -5, k, 4, 5
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__conversion.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..90f45404d094d242761f1140a3c433a3284ac69a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__conversion.txt
@@ -0,0 +1,20 @@
+Quantos(as) segundos estão em 5/3 de um(a) hora?
+6000
+Quantos(as) gramas estão em 1/40 de um(a) tonelada?
+25000
+Quantos(as) meses estão em 0.4160668 anos?
+4.9928016
+Quanto é 2/25 de um(a) milligrama em um(a) microgramas?
+80
+Converte 4102.461l para mililitros.
+4102461
+Quantos(as) meses estão em 4.924495 décadas?
+590.9394
+Quanto é 24/5 de um(a) dia em um(a) minutos?
+6912
+Converte 7392.307 anos para décadas.
+739.2307
+Converte 783.3969 centímetros para metros.
+7.833969
+Quanto é 619964.7t em nanogramas?
+619964700000000000000
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__time.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__time.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4639c3d0d2344961562e1ecc6e0d9602708feb05
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/measurement__time.txt
@@ -0,0 +1,20 @@
+O que é 1 minutos despois de 1:20 AM?
+1:21 AM
+O que é 86 minutos despois de 4:41 AM?
+6:07 AM
+Quantos minutos existem entre 6:26 PM e 12:45 AM?
+379
+O que é 208 minutos despois de 5:09 PM?
+8:37 PM
+O que é 55 minutos despois de 6:43 AM?
+7:38 AM
+O que é 267 minutos despois de 10:26 AM?
+2:53 PM
+O que é 65 minutos antes de 8:50 PM?
+7:45 PM
+O que é 242 minutos despois de 7:05 AM?
+11:07 AM
+O que é 632 minutos antes de 4:14 PM?
+5:42 AM
+Quantos minutos existem entre 5:52 AM e 11:44 AM?
+352
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__base_conversion.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__base_conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..209c3ba27f765d143a84f78abceef952777c2e1c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__base_conversion.txt
@@ -0,0 +1,20 @@
+Converte -113211032 (em base 4) para base 14.
+-272b4
+Quanto é 37996 (base 13) em base 5?
+11241311
+-7661 (em base 15) para base 12
+-1260a
+-143847 (em base 10) para base 9
+-238280
+Converte -13039 (em base 14) para base 8.
+-133153
+Quanto é -35164 (base 15) em base 13?
+-5bc54
+Quanto é -58487 (base 9) em base 16?
+-9880
+155245 (em base 7) para base 2
+111011111000010
+Converte 20173 (em base 14) para base 10.
+77129
+Quanto é 3dac0 (base 14) em base 13?
+5487c
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder.txt
new file mode 100644
index 0000000000000000000000000000000000000000..bbbb7ad9c74915102bd3872b52000154e4834cc3
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder.txt
@@ -0,0 +1,20 @@
+Calcula o resto da divisão entre 83784 e 105.
+99
+Calcula o resto da divisão entre 45012 e 565.
+377
+Calcula o resto da divisão entre 378383 e 54050.
+33
+Qual é o resto da divisão entre 367745 e 133?
+0
+Qual é o resto da divisão entre 30461 e 5065?
+71
+Calcula o resto da divisão entre 534282 e 7973.
+91
+Calcula o resto da divisão entre 483553 e 4605.
+28
+Qual é o resto da divisão entre 5579 e 1031?
+424
+Calcula o resto da divisão entre 32695 e 321.
+274
+Qual é o resto da divisão entre 331609 e 36845?
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6178490bdb44584ae4f213c4553233fa8fd9ba39
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__div_remainder_composed.txt
@@ -0,0 +1,20 @@
+Seja i = 4124 - 3945. Qual é o resto da divisão entre 549 e i?
+12
+Seja o = -4976 + 5185. Calcula o resto da divisão entre o e 6.
+5
+Seja g(y) = 19*y + 15*y - 42*y - 51. Qual é o resto da divisão entre g(-10) e -7*(40/(-14) - -2)?
+5
+Seja w(b) = -3*b - 27. Seja q, w(-9). Qual é o resto da divisão entre (q - 636/(-18))*9/6 e 27?
+26
+Supõe -32*r = -1840 - 4784. Qual é o resto da divisão entre r e 97?
+13
+Supõe p = 5*q - 1202 + 197, -3*p = 0. Qual é o resto da divisão entre q e 26 + (-1 - -2) - (41 - 43)?
+27
+Assume que s é ((-6)/4)/((-6)/16). Supõe -10*c + 5512 = 5472. Supõe -3*z + 3 = -s*z, -5*z = -c*d + 91. Qual é o resto da divisão entre d e 6?
+1
+Supõe -2*h = -2*m - 3*m + 46, 5*h - 60 = -5*m. Supõe -135 = -5*p - 5*x, 85 = m*p - 7*p + 5*x. Qual é o resto da divisão entre 34 e p?
+9
+Supõe 4*j + f - 226 = 11, 4*j - 235 = f. Calcula o resto da divisão entre 170 e j.
+52
+Seja j(u) = 4 - 11*u**2 - u + u**3 + 6*u + 8*u - 21. Seja r(a) = -a**3 + 4*a**2 + 6*a + 5. Seja b, r(5). Calcula o resto da divisão entre 46 e j(b).
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fae4df0a54168bfd27868e4185e9378299a46b99
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd.txt
@@ -0,0 +1,20 @@
+Qual é o maior fator comum  de 69921 e 2839?
+17
+Calcula o maior divisor comum  de 36180 e 460.
+20
+Calcula o maior divisor comum  de 5258 e 36586.
+22
+Calcula o maior divisor comum  de 8298 e 5835338.
+922
+Qual é o maior fator comum  de 125218 e 269630?
+914
+Qual é o maior divisor comum  de 10608472 e 336?
+56
+Calcula o maior divisor comum  de 1441 e 1975873.
+131
+Qual é o maior divisor comum  de 162325 e 72025?
+1075
+Calcula o maior fator comum  de 54 e 270810.
+54
+Calcula o maior divisor comum  de 501 e 45249.
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8184cd2061bdb28ab08257a6993f640aac5feefc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__gcd_composed.txt
@@ -0,0 +1,20 @@
+Assume que a é (-2)/(-4) - 49/2. Assume que d é 1/((-27)/a - 1). Seja l(z) = z**3 + 31*z**2 + 7*z + 233. Seja t, l(-31). Qual é o maior fator comum  de d e t?
+8
+Seja n(p) = 2*p**3 - p**2 - 1. Seja t, n(2). Supõe -4*i = -t*i + 455. Qual é o maior fator comum  de i e 13?
+13
+Supõe 300 = 15*k - 600. Calcula o maior fator comum  de k e 2240.
+20
+Seja h(n) = n**3 - n**2 - 3*n + 3. Seja d = -232 + 235. Seja j, h(d). Qual é o maior divisor comum  de 44 e j?
+4
+Assume que l é -6*(2 - 5)/6. Supõe f - 2 = l. Supõe -5*c - 680 = -f*m, 272 = 2*m + 3*c - 0*c. Calcula o maior divisor comum  de 34 e m.
+34
+Supõe 0 = -13*v - 169*v + 3673 + 474077. Calcula o maior fator comum  de 30 e v.
+15
+Seja d(c) = 17*c**3 + c**2 - 4*c + 2. Seja r, d(1). Seja p = -134 - -218. Assume que m é (36/21)/(1/p). Calcula o maior divisor comum  de m e r.
+16
+Assume que i é 2/3 + (-2)/(-6). Assume que z é 6/i + 2/(-2). Assume que f é (2345/(-134))/(1*2/(-4)). Qual é o maior divisor comum  de f e z?
+5
+Seja p(r) = -5*r**2 + 187*r - 54. Seja l, p(37). Supõe i - 83 = -g, 2*i = 3*g + 4*i - 246. Qual é o maior divisor comum  de l e g?
+20
+Assume que i é 0 - 17*(-3)/3. Supõe -p = r - 31 - i, 0 = r + 4*p - 36. Supõe 0 = 105*g - 755 + 135 - 745. Qual é o maior divisor comum  de r e g?
+13
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor.txt
new file mode 100644
index 0000000000000000000000000000000000000000..db9a1cb7f97dffb8a93b884a5f1dceb2e39dfd82
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor.txt
@@ -0,0 +1,20 @@
+Será que 21 divide 21297129?
+True
+Será 32793800 um múltiplo de 101?
+False
+Será 33543096 um múltiplo de 516?
+True
+Será que 10 divide 12030600?
+True
+Será 59 um fator de 6106544?
+False
+Será 37046561 um múltiplo de 19?
+True
+Será que 12 divide 1992140?
+False
+Será 6 um fator de 2613817?
+False
+Será que 72 divide 4783248?
+True
+Será 56 um fator de 72369976?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9eadd29b37849ca4e0a455de78490f7bf7a67bb1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_factor_composed.txt
@@ -0,0 +1,20 @@
+Supõe 0*o - 39*o + 80109 = 30*o. Será 43 um fator de o?
+True
+Supõe 0 = 5*k - 2*z - 1665, 2*z = k - 6*k + 1685. Seja d = k - 211. Será 11 um fator de d?
+False
+Será 60 um fator de 41*126 + (-5 + 12 - 13)?
+True
+Assume que f é 62 + 4 - (2 + 2). Supõe 0 = 2*c - c - f. Seja t = -24 + c. Será 22 um fator de t?
+False
+Seja r(b) = 51*b**2 + 2*b + 21. Seja v, r(-5). Supõe 6*o = 3*m + 4*o - 770, -4*o = -5*m + v. Será 21 um fator de m?
+False
+Seja j = 32 + 3411. Será que 61 divide j?
+False
+Supõe 4*y + 5*b - 3 = 13, 5*y = 4*b + 20. Seja c(o) = 2*o**2 - 42*o - 21. Seja w, c(25). Supõe 313 = y*v - w. Será 19 um fator de v?
+False
+Seja j(r) = -154*r**3 + 3*r**2 + 2*r. Seja w, j(-1). Supõe -4*u + w + 41 = 0. Será u um múltiplo de 6?
+False
+Supõe 556 = -4*z - l, -4*l - 229 = 3*z + 188. Seja j = z - -191. Será j um múltiplo de 52?
+True
+Supõe 49*o = 2*o + 4576 - 346. Será 18 um fator de o?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4ff4a288de59ceee823b5acc9b04f51a164f47f5
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime.txt
@@ -0,0 +1,20 @@
+Será 1787261185 primo?
+False
+Será 3582758659 um número primo?
+True
+Será 730776089 compósito?
+False
+Será 461689093 primo?
+False
+Será 84231019 primo?
+True
+Será 564826859 primo?
+True
+Será 727429379 um número compósito?
+False
+Será 295283971 um número compósito?
+False
+Será 85278461 um número compósito?
+False
+Será 1457577139 um número compósito?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f0900cb4f78a4a32eda34f4b09a4cc4436038bd1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__is_prime_composed.txt
@@ -0,0 +1,20 @@
+Seja r(p) a terceira derivada de -657*p**4/8 + 4*p**3/3 + 33*p**2. Será r(-1) primo?
+True
+Supõe 0 = -5*z + 10 + 5. Assume que d é (3/(-6))/((z/18)/1). Seja j(c) = 71*c**2 + 3*c + 1. Será j(d) compósito?
+False
+Supõe -3*g - 3*u = -5*g + 20, 4*u = -4*g. Supõe 43 = -0*v + v - g*s, 3*v - 197 = -5*s. Será v um número primo?
+True
+Assume que a é -4 - (-5 + 45/3). Será 7 + -1 - 71974/a compósito?
+False
+Supõe 4*s - 13965 = p, -20*s = -19*s + p - 3490. Será s primo?
+True
+Supõe -h + 3*l = -4*h, 3*l + 21 = 4*h. Supõe h*c - 12 = c - 4*n, 5*n + 7 = 3*c. Supõe 2472 = c*k - 4*a, -4*a - 1858 = -0*k - 3*k. Será k compósito?
+True
+Supõe -2*v + 677 = 667. Seja z(u) = 3*u + 3*u**3 + 4 + 9*u**3 + 4 - 2*u**2. Será z(v) compósito?
+True
+Seja l = -29 + 32. Supõe -2*u + l = 3. Será ((3 - -1) + 19)*(17 + u) um número primo?
+False
+Seja c = -194 + 197. Supõe -1240 + 19791 = w + 2*d, 18550 = w + c*d. Será w um número compósito?
+False
+Seja m(n) a primeira derivada de 127*n**4/2 - n**3/3 + n**2 - 2*n + 11. Seja f, m(1). Supõe 0 = 5*i - f - 3202. Será i um número compósito?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ee21614ad7fc72529ecd774c3a297f12c3184994
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm.txt
@@ -0,0 +1,20 @@
+Calcula o menor múltiplo comum de 598 e 30753.
+18390294
+Calcula o denominador comum de -62/585005 e -13/86479.
+9945085
+Qual é o denominador comum de 83/29565 e -133/78300?
+17147700
+Calcula o denominador comum de -31/3180 e -115/666.
+352980
+Qual é o menor múltiplo comum de 2121350 e 798?
+6364050
+Calcula o denominador comum de 73/199556 e -49/3363944.
+23547608
+Calcula o menor múltiplo comum de 18 e 5250378.
+15751134
+Encontra o denominador comum de -16/33 e -41/595341.
+6548751
+Qual é o menor múltiplo comum de 343665 e 420035?
+3780315
+Calcula o menor múltiplo comum de 4396 e 83524.
+83524
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..aac7c2275bf20a907687f8ed82e91becbb26e7c4
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__lcm_composed.txt
@@ -0,0 +1,20 @@
+Encontra o denominador comum de 9/(-15) + 3562/520 e -55/5096.
+5096
+Supõe 3*b - 1 = 122. Seja w = -29 + b. Supõe 147*d - 48 = 144*d. Qual é o menor múltiplo comum de w e d?
+48
+Calcula o denominador comum de -31/1113 e (-44)/3 - 22/88*24.
+1113
+Calcula o denominador comum de (3776/(-5760))/(2 + 0 - -1) e -37/891.
+8910
+Seja x(s) = -142*s - 1958. Calcula o menor múltiplo comum de 195 e x(-14).
+390
+Qual é o denominador comum de -173/60 e ((4 - 2)/512)/((-43)/(-430))?
+1920
+Seja o(s) = 2*s + 21. Seja d, o(-9). Assume que v é d*(-2 + 3) - -30. Qual é o menor múltiplo comum de 3 e 1/(135/v - 3 - 1)?
+33
+Supõe -9*l = 535 - 1491 - 2482. Calcula o menor múltiplo comum de 4 e l.
+764
+Assume que v é (-12)/18 - (-16)/6. Assume que h é (-6)/12 + 21/v. Supõe -m - h = -2*y + 1, m = -5*y + 45. Calcula o menor múltiplo comum de 2 e y.
+8
+Encontra o denominador comum de 65/38 e 153150/(-10300) + 13 + 2.
+3914
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors.txt
new file mode 100644
index 0000000000000000000000000000000000000000..368c6edb3f810aeedc2a69f25cca050ae5607d08
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors.txt
@@ -0,0 +1,20 @@
+Enumera os fatores primos de 75356763.
+3, 23, 1092127
+Quais são os fatores primos 4553559?
+3, 19, 31, 859
+Enumera os fatores primos de 21867297.
+3, 7289099
+Enumera os fatores primos de 87629182.
+2, 43814591
+Enumera os fatores primos de 23015739.
+3, 17, 37, 12197
+Enumera os fatores primos de 29521083.
+3, 31, 317431
+Enumera os fatores primos de 25843322.
+2, 97, 133213
+Quais são os fatores primos 50800505?
+5, 7, 17, 12197
+Quais são os fatores primos 46101815?
+5, 37, 249199
+Quais são os fatores primos 21609521?
+43, 197, 2551
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ee214d3942affd02cc14e2fa1c8c5efda3c536e4
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__list_prime_factors_composed.txt
@@ -0,0 +1,20 @@
+Quais são os fatores primos (-912)/(-114) - (-15700)/2?
+2, 3929
+Seja d = 9705 + -2895. Enumera os fatores primos de d.
+2, 3, 5, 227
+Supõe 11809 = 24*j - 6359. Supõe -5*x + j = -1058. Quais são os fatores primos x?
+3, 11
+Seja m(i) = -i - 5. Seja r, m(-7). Supõe -4*z + 512 = 3*n, -r*z - n + 92 + 166 = 0. Enumera os fatores primos de z.
+131
+Seja t(u) = -191*u - 58 + 46 - 120 + 182*u. Quais são os fatores primos t(-23)?
+3, 5
+Quais são os fatores primos 144/(-264) - 63570/(-22)?
+3, 107
+Seja a(x) = 30*x**2 - 48*x + 877. Quais são os fatores primos a(18)?
+9733
+Supõe 24*q + 4 = 23*q + 3*r, 0 = 2*q + r - 6. Supõe 5 = -3*l - 7, 3*p = 3*l + 51. Supõe -w + p = c, -q*c + 4 = w - 6. Enumera os fatores primos de w.
+2
+Assume que z é 3/(-15) + (-5272)/(-10). Supõe 297 = 5*i - 2*a - 365, 4*i = -a + z. Enumera os fatores primos de i.
+2, 3, 11
+Seja p(l) = -l**2 + 9*l + 10. Seja r, p(10). Supõe 2*j = 3*j + 2*n - 402, 4*n + 4 = r. Enumera os fatores primos de j.
+2, 101
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0a7886978925eff73ae5a2190639762c36189e07
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value.txt
@@ -0,0 +1,20 @@
+Qual é o dígito das milhões de 76506059?
+6
+Qual é o dígito das centenas de 92525637?
+6
+Qual é o dígito das milhares de 86264362?
+4
+Qual é o dígito das unidades de 9014123?
+3
+Qual é o dígito das dezena milhões de 18850520?
+1
+Qual é o dígito das unidades de 48501266?
+6
+Qual é o dígito das centena milhares de 15351013?
+3
+Qual é o dígito das milhões de 78689710?
+8
+Qual é o dígito das milhares de 55764147?
+4
+Qual é o dígito das centena milhares de 88759692?
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e4ecace3f3ca996cec60d2e3c71b34ffdf87401c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__place_value_composed.txt
@@ -0,0 +1,20 @@
+Qual é o dígito das centenas de ((-31)/(-5))/(5/7775)?
+6
+Seja q(v) = 37*v**3 - 12*v + 4. Seja s, q(4). Supõe 3382 = 18*m - s. Qual é o dígito das unidades de m?
+7
+Seja z = -37 + 73. Assume que t é (4/6)/(12/z). Supõe 0 = 2*b + 4*w - 92 - 4, t*b - w = 106. Qual é o dígito das unidades de b?
+2
+Supõe -n + 0*w - 4*w - 2 = 0, 0 = 3*n - w - 20. Supõe 0 = 4*r + 12, 3*s - n*r + 4*r = 228. Qual é o dígito das dezenas de s?
+7
+Assume que u é (-13)/(-4) - (-2)/((-24)/3). Supõe u*i = 127 - 49. Qual é o dígito das dezenas de i?
+2
+Seja j(v) = -5*v**2 - 3*v - 432. Seja a(d) = -2*d**2 - d - 216. Seja g(n) = -9*a(n) + 4*j(n). Qual é o dígito das unidades de g(0)?
+6
+Supõe -9*t + 8*t = -238. Seja a = t + -18. Qual é o dígito das dezenas de a?
+2
+Seja u(l) = -l**3 - 2*l**2 + 2*l - 1. Seja o, u(0). Qual é o dígito das unidades de 83 - (o + 0)*3?
+6
+Supõe -4*g = 2*c - 6, 0 = -3*c - 0*c - 2*g - 11. Qual é o dígito das dezenas de (-3217)/c + 36/(-63)?
+5
+Qual é o dígito das centenas de 50307/6 - (-20)/(-8)*(-9)/(-15)?
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d9c246a4d33154e22ea27fa9c8b4bf5a8306e2b7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number.txt
@@ -0,0 +1,20 @@
+Arredonda 235711 para as 100000 mais próxima.
+200000
+Quanto é -38336.966 arredondado para as um milhar mais próxima?
+-38000
+Quanto é 8822.777 arredondado para as 10 mais próxima?
+8820
+Arredonda -189.731309 para as casa decimal das doiss.
+-189.73
+Quanto é 0.5699581 arredondado para as casa decimal das 4s?
+0.57
+Arredonda -0.0000188379792 para as casa decimal das 7s.
+-0.0000188
+Arredonda -0.0000134900893 para as casa decimal das seiss.
+-0.000013
+Quanto é 32.5056379 arredondado para as casa decimal das um?
+32.5
+Arredonda 1658337.36 para as 100 mais próxima.
+1658300
+Arredonda -2189.05553 para as 10 mais próxima.
+-2190
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b570afbd563c9b7c2fb77fdf57757d1b2de2a52f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/numbers__round_number_composed.txt
@@ -0,0 +1,20 @@
+Seja t = 116942170670204 + -116942177881929.4799841. Seja o = -7211726 - t. Seja d = o - -0.52. Quanto é d arredondado para as casa decimal das 6s?
+-0.000016
+Seja s = -79.3252 - 0.0748. Seja w = s - -60. Quanto é w arredondado para as casa decimal das zeros?
+-19
+Seja g = -1355511 + -16865083. Seja k = 18220633.999805 + g. Seja y = -40 + k. Quanto é y arredondado para as casa decimal das 5s?
+-0.0002
+Seja n = -783 - -799.475. Seja g = n - 3.189. Seja h = g + -13.5. Quanto é h arredondado para as casa decimal das 2s?
+-0.21
+Seja i = 6726961 - 2535961. Quanto é i arredondado para as um milhão mais próxima?
+4000000
+Seja c = -0.0252 - 45.0748. Seja d = -48 - c. Seja z = d + 7.7. Quanto é z arredondado para as casa decimal das zeros?
+5
+Seja t = -0.4613 - -0.0749. Arredonda t para as casa decimal das 1.
+-0.4
+Seja r = 1.00231 - 209.99031. Seja i = r + 209. Arredonda i para as casa decimal das doiss.
+0.01
+Seja k(i) = -i**3 + 3*i**2 + 24*i + 42. Seja l(f) = f**2 + 8*f + 14. Seja r(w) = 2*k(w) - 7*l(w). Seja u, r(-9). Arredonda u para as 100 mais próxima.
+1400
+Seja b = -0.67 + 0.85. Seja c = b + -0.0755. Arredonda c para as casa decimal das 2s.
+0.1
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__add.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__add.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b53c4d42510428413ab05b6a0a9bb8401ba229d2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__add.txt
@@ -0,0 +1,20 @@
+Seja d(h) = 110*h**2 + 14*h - 14. Seja n(z) = 18*z + 141. Seja x, n(-8). Seja j(w) = 22*w**2 + 3*w - 3. Determina x*d(c) + 14*j(c).
+-22*c**2
+Seja z(i) = -i. Seja r(a) a terceira derivada de a**4/6 - a**3 - 108*a**2 - 2. Obtém -r(y) - 5*z(y).
+y + 6
+Supõe -9*m + 10*m - 2*o = -9, -4*m = -4*o + 24. Seja f(d) = 7*d**3 - 2*d**2 + 3*d + 1. Seja h(i) = -8*i**3 + 3*i**2 - 4*i - 2. Calcula m*f(n) - 2*h(n).
+-5*n**3 - n + 1
+Seja c(j) = 126*j**2 + 21*j + 23. Seja d(l) = 43*l**2 + 6*l + 8. Determina -2*c(a) + 7*d(a).
+49*a**2 + 10
+Seja b(n) a segunda derivada de -n**3/6 + 2*n**2 + 2*n. Seja k(l) a segunda derivada de l**3/3 - 3*l**2 - l. Supõe 4*u + 28 - 38 = -42. Obtém u*b(i) - 5*k(i).
+-2*i - 2
+Seja d(j) = 32*j**2 + 3*j - 3. Seja g(n) = -2465*n**2 - 232*n + 232. Supõe 268*k + 3 = 269*k - 3*a, 5*k = -3*a + 15. Obtém k*g(f) + 232*d(f).
+29*f**2
+Seja g(u) = 1 - 2*u**2. Seja d(j) = -12*j**2 + j + 9. Calcula -d(o) + 5*g(o).
+2*o**2 - o - 4
+Seja m(p) a terceira derivada de -p**4/6 + 89*p**2 - 2. Seja g(i) = 1 - 4*i. Determina -3*g(v) + 4*m(v).
+-4*v - 3
+Seja k(s) = 5*s**2 - 12*s + 4. Seja c(f) = f**2 - 2*f. Seja z(n) = 4*c(n) - k(n). Seja i(g) = -2*g**2 + 4*g**2 + 27*g + 3 - 30*g. Obtém 4*i(b) + 3*z(b).
+5*b**2
+Seja v(u) = 8*u + 3. Seja l(y) = 3*y + 1. Seja n(a) = 11*l(a) - 4*v(a). Seja x(d) = 2. Obtém -3*n(t) - x(t).
+1 - 3*t
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__coefficient_named.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__coefficient_named.txt
new file mode 100644
index 0000000000000000000000000000000000000000..77c8fe967e0f5c9620da09eaaf5c498d76c762f9
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__coefficient_named.txt
@@ -0,0 +1,20 @@
+Expressa 7*b + 23 - 54*b**2 + 218*b**2 - 83*b**2 - 76*b**2 na forma i*b + p + d*b**2 e dá d.
+5
+Rearranja (-6 + 4*s**2 + 6)*(-3*s**2 + 3*s**2 - 2*s**2) - 8945*s + 8945*s - 985*s**4 como c*s**2 + p*s**3 + v*s + q + d*s**4 e dá d.
+-993
+Expressa 0*x + 13685*x**3 + 2*x - 3*x**2 - 13684*x**3 - 21 - 2*x**4 como g*x**3 + r*x**2 + t*x + b*x**4 + c e dá t.
+2
+Expressa 256*n**2 - 127*n**2 - 130*n**2 - 77*n + 1137*n - 1 como m*n + d*n**2 + f e dá m.
+1060
+Rearranja (-20 + 71 + 70 + (1 + 3 - 5)*(5 + 2 - 5))*(-24*m**4 + 26*m**4 + 16*m**4) na forma g + q*m**4 + u*m**3 + j*m**2 + z*m e dá q.
+2142
+Expressa (-2 + 6 + 2)*(2 + 2 - 5)*(1149*r - 1575*r + 880*r) como w*r + g e dá w.
+-2724
+Expressa 33*x**4 + 181 - 181 + (-x**2 - 4*x + 4*x)*(x**2 + 3 - 3) - 898*x**2 + 8*x**4 + 898*x**2 na forma h + j*x + t*x**4 + g*x**2 + c*x**3 e dá t.
+40
+Rearranja (-w**2 - 2*w + 2*w)*(3*w - w + 4*w) + (6 - 16 - 4)*(-3*w + 0 + 0)*(-9*w + 4*w**2 + 3*w - 3*w**2) como n*w + s + y*w**2 + z*w**3 e dá z.
+36
+Rearranja (-3 + 3 + 3*t)*(1 - 608*t - 8810*t**2 + 4432*t**2 + 4376*t**2) na forma g*t + s*t**2 + r*t**3 + d e dá g.
+3
+Expressa (7 - 1 + 0 + (-69 - 86 + 141)*(0 + 2 + 0))*(t**4 + 4 - 4) como u*t**4 + v*t + s + w*t**3 + y*t**2 e dá u.
+-22
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__collect.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__collect.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5390d5f85a74849852dcdac421f510a9f6edb1bc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__collect.txt
@@ -0,0 +1,20 @@
+Coleta os termos em 3*a + 20*a - 23*a - 28018*a**2.
+-28018*a**2
+Coleta os termos em -6 - 15518*x + 15643*x + 19 + 5.
+125*x + 18
+Coleta os termos em 300*u + 3 - 1 - 264*u - 2 + 0.
+36*u
+Coleta os termos em -4*r + r + 18171*r**3 - 301788*r**3.
+-283617*r**3 - 3*r
+Coleta os termos em -11825*f**2 + 5749*f**2 + 5904*f**2 - 3*f.
+-172*f**2 - 3*f
+Coleta os termos em -2720*v + 1072*v - 3727*v + 104*v.
+-5271*v
+Coleta os termos em -44*w**3 + 14*w + 43*w**2 + 17*w - 33*w**2 - 32*w.
+-44*w**3 + 10*w**2 - w
+Coleta os termos em -26 + 64 - 30 - 24*b**2 + 19*b**2.
+8 - 5*b**2
+Coleta os termos em 324*l**2 - 146*l**2 - 148*l**2 - 8*l + 3*l - 20*l**3.
+-20*l**3 + 30*l**2 - 5*l
+Coleta os termos em -15*w**3 - 11*w**3 - 12*w**3 + 69*w**3 + 2*w**2 - 15*w**3.
+16*w**3 + 2*w**2
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__compose.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__compose.txt
new file mode 100644
index 0000000000000000000000000000000000000000..43d1ea9f3578bb6ded648af3be272b8461b7e66b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__compose.txt
@@ -0,0 +1,20 @@
+Seja t(z) = z + 32. Seja d(w) = -w**2 - w - 32. Seja i(j) = -4*d(j) - 4*t(j). Seja c(s) = 2 - 42*s. O que é c(i(k))?
+2 - 168*k**2
+Seja s(t) = 2666*t - 5337*t + 125*t**2 + 2671*t. Seja m(a) = a**2 + a - 1. Seja x(o) = -10*o**2 - 12*o + 12. Seja w(c) = -12*m(c) - x(c). Obtém s(w(r)).
+500*r**4
+Seja i(n) a terceira derivada de -13*n**5/30 + 15652*n**2. Seja v(j) = -j - 2. Seja y(t) = -3*t - 9. Seja h(d) = 9*v(d) - 2*y(d). Obtém i(h(m)).
+-234*m**2
+Seja y(i) = 141*i. Seja l(u) = 30442*u. O que é l(y(m))?
+4292322*m
+Seja l(m) = -25351*m**2. Seja x(y) = -y**2 - 133*y. Determina x(l(n)).
+-642673201*n**4 + 3371683*n**2
+Seja y(i) = -4 + 10 + 5 - 35*i - 1 - 7. Seja z(h) = -8*h. Determina z(y(v)).
+280*v - 24
+Seja n(h) = 1587*h. Seja k(z) = 28*z - 52. Seja s(t) = 13*t - 24. Seja y(j) = -6*k(j) + 13*s(j). O que é y(n(l))?
+1587*l
+Seja p(d) = -2*d**2. Seja j(x) a segunda derivada de 8/3*x**3 + 49*x + 0 + 0*x**2. Calcula p(j(k)).
+-512*k**2
+Seja v(b) = 2*b + 2253. Seja m(g) = -4*g - 42. Determina m(v(z)).
+-8*z - 9054
+Seja h(s) a primeira derivada de 2*s**3 + 10*s - 869. Seja k(g) = 3*g**2. Determina h(k(c)).
+54*c**4 + 10
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a6b531ae80362eb54ba3b62634f844e2b1ee0223
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate.txt
@@ -0,0 +1,20 @@
+Seja i(c) = 481940 - 2802*c. Calcula i(172).
+-4
+Seja o(c) = -416*c - 435. Calcula o(-2).
+397
+Seja a(u) = 2*u**3 - 62*u**2 - 441*u - 342. Determina a(37).
+-231
+Seja n(g) = 257*g - 6195. Determina n(24).
+-27
+Seja h(g) = -g**2 - 37*g + 11359. Calcula h(90).
+-71
+Seja l(o) = -3*o**2 + 434*o - 2485. Calcula l(6).
+11
+Seja v(w) = -7892*w - 31433. Calcula v(-4).
+135
+Seja r(c) = -33992*c - 169967. O que é r(-5)?
+-7
+Seja t(y) = y**3 + 6*y**2 - 702*y - 415. O que é t(24)?
+17
+Seja z(j) = 36*j**2 - 3671*j - 205. O que é z(102)?
+-103
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9169e882ecc48d035eaa0bce4a9ae5c9c0617153
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__evaluate_composed.txt
@@ -0,0 +1,20 @@
+Seja r(d) a segunda derivada de -d**8/6720 + d**7/630 + d**6/360 + d**5/120 + d**4/3 - d**2/2 - 47*d. Seja k(j) a terceira derivada de r(j). Calcula k(4).
+9
+Seja i(d) = d**3 - 6*d**2 + 4*d + 2. Seja p, i(5). Seja o(h) = h + 1. Seja q(c) = -c - 2. Seja x(y) = p*q(y) - 5*o(y). Seja n = 16 + -14. Obtém x(n).
+-3
+Seja o(x) a segunda derivada de -1/6*x**3 + x + x**2 - 18. Obtém o(-6).
+8
+Seja f = 35 + -38. Assume que n é 1*-7*f/(2 + -5). Seja z(p) = p**2 + 4*p + 1. Seja x(m) = -m - 1. Seja o(s) = -3*x(s) + z(s). Determina o(n).
+4
+Seja b(z) = 2*z**2 - 33*z + 16. Seja s(t) = 4*t**2 - 65*t + 32. Seja k(q) = 7*b(q) - 4*s(q). Seja p, k(14). Seja a(u) = -5*u**2 - u + 1. Calcula a(p).
+-17
+Seja t(u) = 67*u**3 + 12*u**2 - 5*u - 2. Seja o(i) = -25*i**3 - 4*i**2 + 2*i. Seja m(q) = -8*o(q) - 3*t(q). Determina m(-3).
+0
+Seja g(w) a terceira derivada de w**8/6720 - w**7/840 + w**6/720 + w**5/60 + 11*w**4/6 + w**3/6 - 50*w**2. Seja n(y) a segunda derivada de g(y). Obtém n(2).
+0
+Seja i(l) = -l**3 + 20*l**2 + 2*l - 39. Seja m(k) = 2*k + 32. Seja z, m(-6). Seja n, i(z). Seja a(v) = -7*v**3 - v**2 + 1. Determina a(n).
+-7
+Seja i(t) = t + 4. Seja j = -127 - -130. Assume que q é 1/j - (-351)/(-81). O que é i(q)?
+0
+Supõe -47*y = -37*y - 140. Seja m(s) = -y + 14 + s - 4. O que é m(7)?
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__expand.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__expand.txt
new file mode 100644
index 0000000000000000000000000000000000000000..97e1494a5599815549532473862945c038323b31
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__expand.txt
@@ -0,0 +1,20 @@
+Expande -m - m**4 - 52*m**2 + 52*m**2 + (49*m - 49*m - 6*m**4 + (-m**2 + 0*m**2 + 3*m**2)*(2*m**2 - 3*m**2 - m**2))*(-11 + 13 + 10).
+-121*m**4 - m
+Expande (-3*r**2 - 6*r**2 - 5*r**2)*(-239 - 263 + 188).
+4396*r**2
+Expande (u + 4*u - u)*(-43989*u**3 - 1 - 41*u**4 + 43987*u**3 - 294*u**4).
+-1340*u**5 - 8*u**4 - 4*u
+Expande (-b - 4*b - 2*b)*(441 - 441 + 123*b).
+-861*b**2
+Expande (0*h - 3*h - 127*h)*(-7*h - 7*h + 9*h).
+650*h**2
+Expande -4*d + d + 5*d + (-10 + 13 + 28)*(-4*d + d + 2*d + (-1 - 2 + 5)*(2 + 2*d - 2)).
+95*d
+Expande (2*g**3 + 3*g**3 - 3*g**3)*(6 - 3 + 3)*(-89 + 6 + 51).
+-384*g**3
+Expande (8*o - o - 2*o)*(314 - 1057 - 678).
+-7105*o
+Expande (2 + 4 - 5)*(2*d**2 + 2*d - 2*d) + (-226*d - 151*d + 300*d)*(-d + 6 - 6).
+79*d**2
+Expande (3 - 1 + 0)*(-4*w**2 + 6*w**2 - w**2) + (18*w - 14*w + 20*w)*(0*w + 3*w - 8*w) + 0*w - 2*w**2 + 0*w.
+-120*w**2
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__simplify_power.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__simplify_power.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3deb7dfe4170824b089ba3f58dc57f625e8bf204
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/polynomials__simplify_power.txt
@@ -0,0 +1,20 @@
+Simplifica (u**(1/8))**16/(u**(-2/15)/((u**(-1/4)*u*u)/u*u*u))**(-2/59) assumindo que u é positiva.
+u**(3367/1770)
+Simplifica ((y**(2/5)/y)/(y*y**(-4/7)/y*y))**12*(y**(-2/3)*((y*y**2*y)/y)/y)/((y**(-4)*y)/(y/(y/((y/(y/y**1))/y*y)))) assumindo que y é positiva.
+y**(-736/105)
+Simplifica ((a/a**(1/3))/a**3)**(-36/7)/(a**(4/9)/(a**1/a))**(17/5) assumindo que a é positiva.
+a**(472/45)
+Simplifica ((s*s**(5/2)*s**(2/13))/((s*s**(-1)*s)/s**(-1/4)))/(((s*s*s**(3/4))/(s*(s*s/(s/(s**1*s)))/s*s))/s**(-15)) assumindo que s é positiva.
+s**(-295/26)
+Simplifica i**26/(i*i**(2/3)*i**(-3)*i)*(i**31)**(6/19) assumindo que i é positiva.
+i**(2059/57)
+Simplifica (((m*m**(4/5)/m*m)/m)/(m**4*m))/m**(2/111)*((m*m**(-4/7)/m*m)/m**4)**(16/7) assumindo que m é positiva.
+m**(-336709/27195)
+Simplifica ((u/((u*u/(u/((u*u**(-2))/u*u)))/u))**(1/25)/((u/u**(9/7))/u**(-1/2)))**(-2/159) assumindo que u é positiva.
+u**(47/27825)
+Simplifica (n**4/(n/(n*(n*n/(n**(-3)*n*n))/n*n)))**(-2/49)/((((n*n/(n/n**(1/2))*n)/n)/n*n**0)/n**22) assumindo que n é positiva.
+n**(297/14)
+Simplifica ((n/n**(1/7))**(2/115)*(n**(3/2))**(19/3))**(-2/39) assumindo que n é positiva.
+n**(-15319/31395)
+Simplifica ((g**(-5))**(10/11)*(g*g**4*g)/g*g**4)**23 assumindo que g é positiva.
+g**(1127/11)
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_level_set.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_level_set.txt
new file mode 100644
index 0000000000000000000000000000000000000000..219816fb80468a6be4cac804b04b8cdedbb95350
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_level_set.txt
@@ -0,0 +1,20 @@
+Dois caractéres são escolhidos sem reposição de tkytenytekn. Dá a probabilidade de(a) escolher 2 k.
+1/55
+Qual é a probabilidade de(a) escolher 1 d e 2 y quando três caractéres são escolhidos sem reposição de yddygyeyygecddyeyyee?
+28/285
+Calcula da probabilidade de(a) escolher 1 z e 1 o quando dois caractéres são escolhidos sem reposição de {d: 1, u: 1, k: 1, o: 1, z: 2}.
+2/15
+Calcula da probabilidade de(a) escolher 2 v e 1 f quando três caractéres são escolhidos sem reposição de {b: 1, f: 1, a: 2, v: 1, x: 6, k: 3}.
+0
+Três caractéres são escolhidos sem reposição de cgatccatr. Dá a probabilidade de(a) escolher 1 a e 2 c.
+1/14
+Três caractéres são escolhidos sem reposição de {y: 3, r: 2, c: 1, x: 2, k: 1}. Qual é a probabilidade de(a) escolher 1 c e 2 r?
+1/84
+Quatro caractéres são escolhidos sem reposição de wqaaaafa. Qual é a probabilidade de(a) escolher 2 a, 1 f e 1 w?
+1/7
+Dois caractéres são escolhidos sem reposição de cfcaofcocobocc. Dá a probabilidade de(a) escolher 1 b e 1 o.
+4/91
+Três caractéres são escolhidos sem reposição de {v: 7, h: 3, d: 3, t: 3}. Dá a probabilidade de(a) escolher 1 d e 2 t.
+9/560
+Quatro caractéres são escolhidos sem reposição de vvcvvvccvcc. Qual é a probabilidade de(a) escolher 4 v?
+1/22
diff --git a/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_sequence.txt b/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_sequence.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0e0c027528597291279fc7c038502caccbcfcb82
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/interpolate/probability__swr_p_sequence.txt
@@ -0,0 +1,20 @@
+Dois caractéres são escolhidos sem reposição de lcztclxzpcpcxxzzzc. Dá a probabilidade de(a) sequência lz.
+5/153
+Dois caractéres são escolhidos sem reposição de bbtbbb. Qual é a probabilidade de(a) sequência bt?
+1/6
+Três caractéres são escolhidos sem reposição de {l: 3, y: 4, k: 3}. Qual é a probabilidade de(a) sequência ykk?
+1/30
+Qual é a probabilidade de(a) sequência ege quando três caractéres são escolhidos sem reposição de eeegeeeegegeegggg?
+21/136
+Quatro caractéres são escolhidos sem reposição de {m: 7, k: 11}. Qual é a probabilidade de(a) sequência kmkm?
+77/1224
+Calcula da probabilidade de(a) sequência bxxa quando quatro caractéres são escolhidos sem reposição de xaamabmmmaaxamam.
+1/3120
+Três caractéres são escolhidos sem reposição de aewgeeezege. Qual é a probabilidade de(a) sequência eza?
+1/165
+Quatro caractéres são escolhidos sem reposição de {d: 6, q: 14}. Qual é a probabilidade de(a) sequência qqdq?
+182/1615
+Três caractéres são escolhidos sem reposição de {m: 1, e: 4, b: 1, q: 1, y: 1, z: 1}. Dá a probabilidade de(a) sequência eyy.
+0
+Quatro caractéres são escolhidos sem reposição de xxxmmmxxmmxmmxx. Qual é a probabilidade de(a) sequência xxmx?
+14/195
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b2f87bcef9a79e81f0424d3ce12aeda85baba485
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d.txt
@@ -0,0 +1,6 @@
+Resolve 9*k = 7*k + 4 para k.
+2
+Resolve 3*y - 27*y - 96 = 0 para y.
+-4
+Resolve -7*x + 3 - 24 = 0 para x.
+-3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..49f8bbf70e76bcb78f5210e96540eb58e08c0ab0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_1d_composed.txt
@@ -0,0 +1,6 @@
+Supõe 0 = j - 4*j - 21. Seja b(d) = d**3 + 7*d**2 + 5. Seja v, b(j). Supõe 40 + 25 = 5*i. Resolve -i = v*q - 38 para q.
+5
+Assume que y é ((-2)/(-1))/((-2)/(-3)). Supõe 3*c - i - 18 + 1 = 0, 0 = 4*c + y*i - 1. Supõe c*f = 3*f + 2. Resolve 0 = 4*p - f*p - 8 para p.
+4
+Assume que m é 10*7/((-14)/(-3)). Resolve 2*u + m = 7*u para u.
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b08a1f7bdf486cc7b8d196211829ae399b67c9cd
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d.txt
@@ -0,0 +1,6 @@
+Resolve 0 = r + 2*g - 8, -5*g = -4*r - 16 - 17 para r.
+-2
+Resolve 4*q - 26*r + 4 = -30*r, -3*q - 3 = -2*r para q.
+-1
+Resolve -x - 25 = -5*q, -q + 19 = -8*x + 5*x para x.
+-5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..52bc281cabe5231de1efc744c893775574303f1c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__linear_2d_composed.txt
@@ -0,0 +1,6 @@
+Supõe 0 = -14*p + 16*p - 6. Supõe -5*x + 6 = 1. Resolve -q - 5 = -3*v - p*q, -3*v + 2*q = -x para v.
+1
+Assume que q é 6*1/(6/4). Supõe -3*t + 8 + q = 0. Resolve n - o = -0*n + 4, 0 = t*n - 2*o - 12 para n.
+2
+Seja f(k) = 4 - k. Supõe -4*s + 7 = -5. Assume que v é ((0 - s) + 1)/(-1). Seja i, f(v). Resolve i*c = d + 4, -3*c - c + 28 = 3*d para c.
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4977f02f633e4244b2dbc8d60679a7ea002d0740
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots.txt
@@ -0,0 +1,6 @@
+Determina q para que -q**3/6 + 5*q**2/6 + 3*q/2 - 15/2 = 0.
+-3, 3, 5
+Seja -4*a**5 - 20*a**4 + 48*a**3 + 176*a**2 - 320*a = 0. Qual é a?
+-5, -4, 0, 2
+Qual é o valor de o em -3*o**4/7 + 15*o**3 - 1287*o**2/7 + 6171*o/7 - 7986/7 = 0?
+2, 11
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..586339590f4cc051af5a656a483d0905f63d0cb0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__polynomial_roots_composed.txt
@@ -0,0 +1,6 @@
+Seja c(k) = 12 - k. Supõe -4*v = -v - 5*y - 42, -4*y = 2*v - 6. Seja j, c(v). Determina q para que 2*q**5 + 2*q**j - q**4 - 2*q**4 - q**4 = 0.
+0, 1
+Seja b(r) = -7*r**2 + 2*r + 5. Seja x(h) a primeira derivada de 0*h**2 - 1/3*h**3 + h + 2. Seja v(j) = b(j) - 5*x(j). Factoriza v(o).
+-2*o*(o - 1)
+Assume que u é 11/(-4) + 2 - -1. Supõe 4*b - 32 = 5*d, -3*b - 3*d + 5 = -2*d. Seja u*n + 0 + n**2 - 3/4*n**5 - n**4 + 1/2*n**b = 0. Calcula n.
+-1, -1/3, 0, 1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_next_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_next_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9602469aa0d444a1345df4b563c8156f0f8900b2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_next_term.txt
@@ -0,0 +1,6 @@
+O que vem a seguir: 105, 210, 315, 420, 525, 630?
+735
+Qual é o termo seguinte em -870, -866, -862, -858, -854, -850?
+-846
+Qual é o termo seguinte em -33, -105, -225, -393, -609, -873?
+-1185
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_nth_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_nth_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..72a84c28468cae08bd23bd29a7af4f3bbf2ed40f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/algebra__sequence_nth_term.txt
@@ -0,0 +1,6 @@
+Qual é termo nº x na sequência -118, -114, -108, -100, -90, -78, -64?
+x**2 + x - 120
+Qual é termo nº z na sequência 2183, 2168, 2151, 2132?
+-z**2 - 12*z + 2196
+Qual é termo nº l na sequência -176, -303, -430, -557, -684, -811?
+-127*l - 49
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5758c4d22041cbe90b84b0ac7de5bb9a6e51575a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub.txt
@@ -0,0 +1,6 @@
+Junta -0.693 e -16.
+-16.693
+Quanto é -1.82 menos -0.2456?
+-1.5744
+0.01526 + 0.027
+0.04226
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub_in_base.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub_in_base.txt
new file mode 100644
index 0000000000000000000000000000000000000000..25e8f607cfdc5213c3686c44bc43ebb125ecfe43
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_or_sub_in_base.txt
@@ -0,0 +1,6 @@
+Em base 10, quanto é -414 - 11?
+-425
+Em base 10, quanto é -4556 + -4?
+-4560
+Em base 6, quanto é -5513 - -15?
+-5454
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_sub_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_sub_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6a57898a4905731f2cbeb6c37b6bfb03d8420a92
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__add_sub_multiple.txt
@@ -0,0 +1,6 @@
+Avalia 9 - (6 + 8 + -6).
+1
+Calcula (5 - -2) + -4 + -3.
+0
+Calcula (10 - 10) + -12 - -5.
+-7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__div.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__div.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7da14153fca4e3f8da600f0a7e4a01678bca9b81
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__div.txt
@@ -0,0 +1,6 @@
+Quanto é -610 dividido por 5?
+-122
+-1 dividido por 8362
+-1/8362
+Quanto é 539 dividido por 11?
+49
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mixed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mixed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ff2c591e10dc02d12ba305188c617971fff3feaf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mixed.txt
@@ -0,0 +1,6 @@
+Avalia 62/84 + (-12)/63*3.
+1/6
+Calcula ((-16)/(-30))/(16/(-6) - -2).
+-4/5
+((-8)/1 + 4)*(-1)/(-2)
+-2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9e9f4433ef6e5ff5729a66d427c6223ddac79b73
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul.txt
@@ -0,0 +1,6 @@
+Multiplica -27 e 0.3.
+-8.1
+Quanto é -0.35 vezes -2?
+0.7
+Quanto é -255 vezes -6?
+1530
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul_div_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul_div_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..09f618ab238b2ffc3583216b172ca4aa677e3109
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__mul_div_multiple.txt
@@ -0,0 +1,6 @@
+Calcula ((-6)/10)/((-264)/352).
+4/5
+0*20/(-8)*2/10
+0
+Quanto é -12*((-2)/(-5))/((-66)/55)?
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__nearest_integer_root.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__nearest_integer_root.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7f11b52849458fbf1d5d6f76fa1829b297f0660e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__nearest_integer_root.txt
@@ -0,0 +1,6 @@
+Quanto é 1093 elevado a 1/7, arredondado para o inteiro mais próximo?
+3
+Qual é a quarta raíz de 811 arredondada para o inteiro mais próximo?
+5
+Quanto é 21605 elevado a 1/2, arredondado para o inteiro mais próximo?
+147
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__simplify_surd.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__simplify_surd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d5499b56794d8d6a47ab96eb6a108b8df2486fa5
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/arithmetic__simplify_surd.txt
@@ -0,0 +1,6 @@
+Simplificar sqrt(2)*1 + -3 - sqrt(2) - -6*sqrt(32).
+-3 + 24*sqrt(2)
+Simplificar -3 + (sqrt(2) - (sqrt(6)/sqrt(3) + sqrt(2)))**2 + sqrt(2) + -3.
+-4 + sqrt(2)
+Simplificar sqrt(27)/(sqrt(576) - sqrt(36)).
+sqrt(3)/6
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6ffbca9358104ae33be86f83aeeea6a1206d7e21
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate.txt
@@ -0,0 +1,6 @@
+Encontra a terceira derivada de 609*s**4 + 2*s**2 + 292 em relação a s.
+14616*s
+Encontra a terceira derivada de 150*o**4 + 8*o**3 + 1830*o**2 em relação a o.
+3600*o + 48
+Encontra a terceira derivada de 650*d**3*h**3 - 2*d**2*h**3 - 102*h**2 em relação a d.
+3900*h**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..102452c264059bdcd44070af862dedba464b44c4
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/calculus__differentiate_composed.txt
@@ -0,0 +1,6 @@
+Seja n(s) = 2*s - 2. Seja o(q) a primeira derivada de 2*q**2 - 5*q + 3. Seja p(u) = 7*n(u) - 3*o(u). Qual é a derivada de p(r) em relação a r?
+2
+Seja d(u) a terceira derivada de -7*u**5/60 - 7*u**3/6 + 40*u**2. Qual é a primeira derivada de d(o) em relação a o?
+-14*o
+Seja f(l) a segunda derivada de 0*l**3 + 1/4*l**4 - 3/20*l**5 + 0 + 0*l**2 - 3*l. Encontra a terceira derivada de f(q) em relação a q.
+-18
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fb2d98724cac45b5dee15f6671a72fe04895ab89
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest.txt
@@ -0,0 +1,6 @@
+Qual valor está mais perto de -39 em -1/3, -1/8, -3?
+-3
+Qual está mais próximo de 2?  (a) 2/9  (b) 13  (c) 1/2
+c
+Qual está mais próximo de 0.1?  (a) -2  (b) 5/6  (c) -4
+b
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..131b92a45b61744849f149bdb1be0ffeb9d4483f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__closest_composed.txt
@@ -0,0 +1,6 @@
+Seja u = 0 + 2. Assume que q é ((-1)/3)/(66/81). Seja m = 12/11 - q. Qual valor está mais perto de 0.1 em u, m, -4?
+m
+Seja r = -179 - -109. Assume que u é 15/3*4/r. Qual valor está mais próximo de 0.1 em u, 2, -0.3?
+u
+Seja v = -5/21 + -2/21. Supõe k - 5*x + 5 = -0*k, k = -3*x + 3. Supõe -8 - 12 = -4*i. Qual valor está mais próximo de k em v, i, 2/5?
+v
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..498c08655481e9de9bc82ff4884d454925dc356d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest.txt
@@ -0,0 +1,6 @@
+Qual é o terceiro maior valor em 0.3, 1/6, -7.19?
+-7.19
+Qual valor é maior?  (a) -0.3  (b) -332  (c) -3
+a
+Qual é o segundo menor valor em 9, 36/5, 4, 0.3?
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f83bb111d2ac9108f23fdeccb80ffe60810a7790
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__kth_biggest_composed.txt
@@ -0,0 +1,6 @@
+Seja g = -1.3 - 2.7. Qual valor é terceiro menor?  (a) g  (b) 2/15  (c) 0.4
+c
+Seja t(k) = -3*k - 4. Seja u, t(-3). Seja y(q) = 2*q**2 + 4*q. Seja z, y(u). Assume que i é (-6)/z + 2/7. Qual valor é segundo menor?  (a) 1/6  (b) -4  (c) i
+a
+Seja w = -1.54 + 1.4. Qual valor é maior?  (a) w  (b) -2/9  (c) -1/3
+a
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4e2e3d939953f1b224d26a1281c7e2389caa94ab
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair.txt
@@ -0,0 +1,6 @@
+Será -15 menor do que -213/13?
+False
+Qual é menor: 218/27 ou 7?
+7
+Qual é menor: 1885/2 ou -0.1?
+-0.1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..47739452ef78df81925c04d0592929aa95341b7c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__pair_composed.txt
@@ -0,0 +1,6 @@
+Seja q = 7 + 1. Assume que o é q/6*(-3)/22. Assume que r é (-3 + 15/3)/(-2). Qual é menor: o ou r?
+r
+Seja m(r) = r**3 + 5*r**2 + 5*r - 10 + 6 - 2*r. Seja c, m(-3). Supõe c*g = 2*z + 4, -3*z - z = 2*g - 16. Será z, no máximo, 2?
+False
+Supõe 0 = 4*z + 4*v + 4, z = 4*z + v + 1. Será -9/7, no máximo, tão grande quanto z?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1c497b6a49731df89f0e5c2c56b5bd2d17afe518
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort.txt
@@ -0,0 +1,6 @@
+Ordena -1.5, 2, 4.
+-1.5, 2, 4
+Coloca -41/3, 2/9, -0.3, 0.4, 3 em ordem crescente.
+-41/3, -0.3, 2/9, 0.4, 3
+Ordena 1/5, -107, 0.5.
+-107, 1/5, 0.5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8ffadb65a96a18760f1addb11fbbd8fded90e0c3
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/comparison__sort_composed.txt
@@ -0,0 +1,6 @@
+Seja m(z) = -z**2 + 4*z + 1. Seja w, m(4). Seja r(o) = -4*o**3 - 2*o - 1. Seja p, r(-1). Supõe 2*v = p*v + 3. Ordena 2, v, w por ordem ascendente.
+v, w, 2
+Seja r(c) = c**3 - 6*c**2 + c - 2. Seja d, r(6). Supõe -3*g + d = 1. Coloca g, -2, 4 em ordem decrescente.
+4, g, -2
+Seja u = 2.6 + -3.5. Ordena u, 4, 5 por ordem ascendente.
+u, 4, 5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..019d468997439c23aaa5847473f82557429494cf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__conversion.txt
@@ -0,0 +1,6 @@
+Quantos(as) micrómetros estão em 1/10 de um(a) metro?
+100000
+Quantos(as) décadas estão em 479.9781 milénios?
+47997.81
+Quanto é 5.716473 litros em mililitros?
+5716.473
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__time.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__time.txt
new file mode 100644
index 0000000000000000000000000000000000000000..39b3b327d796151da05baa20f8ab2de2f3c355e2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/measurement__time.txt
@@ -0,0 +1,6 @@
+O que é 171 minutos despois de 3:12 AM?
+6:03 AM
+Quantos minutos existem entre 2:28 AM e 9:53 AM?
+445
+Quantos minutos existem entre 6:00 AM e 11:24 AM?
+324
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__base_conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__base_conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e1f9ec722c6e5001784ac18b68c12a1ac54e2154
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__base_conversion.txt
@@ -0,0 +1,6 @@
+Quanto é -d (base 15) em base 13?
+-10
+-7 (em base 8) para base 12
+-7
+Converte 7 (em base 12) para base 8.
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2286b1ea54d3158b6ebdbdeabdf8f3ade3c306f0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder.txt
@@ -0,0 +1,6 @@
+Qual é o resto da divisão entre 270 e 50?
+20
+Calcula o resto da divisão entre 217 e 14.
+7
+Calcula o resto da divisão entre 4229 e 90.
+89
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7625eeb8a890a1aec90a9db5f17ab61b396ca574
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__div_remainder_composed.txt
@@ -0,0 +1,6 @@
+Supõe -10 = -5*m + q + 203, -q - 45 = -m. Calcula o resto da divisão entre 125 e m.
+41
+Seja s = 36 - 24. Seja g = s + 8. Seja j = -14 + 21. Qual é o resto da divisão entre g e j?
+6
+Seja g = 91 - 64. Seja u = 16 - g. Seja c = u - -26. Calcula o resto da divisão entre c e 8.
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7835f31d9c1be3886e1ecad768992385c66a08c7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd.txt
@@ -0,0 +1,6 @@
+Qual é o maior fator comum  de 1012 e 736?
+92
+Calcula o maior divisor comum  de 175 e 200.
+25
+Calcula o maior divisor comum  de 5668 e 52.
+52
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2b9ab4c46e30ed2a7dbe6f643c2e73db8390c582
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__gcd_composed.txt
@@ -0,0 +1,6 @@
+Supõe 5*q = -4*u, -2*u - 2*q = 3*u. Supõe u = -5*v - 0*v. Supõe v = -5*b - 3 + 23. Calcula o maior divisor comum  de b e 2.
+2
+Supõe 0 = 24*s - 62 - 106. Qual é o maior fator comum  de s e 2?
+1
+Seja j(q) = 54*q + 16. Seja y, j(5). Calcula o maior fator comum  de 22 e y.
+22
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor.txt
new file mode 100644
index 0000000000000000000000000000000000000000..626f11f4dd3b012d77b87b18c0bace88d5e32e87
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor.txt
@@ -0,0 +1,6 @@
+Será 518 um múltiplo de 37?
+True
+Será que 24 divide 8194?
+False
+Será 18 um fator de 318?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..151a3146044d15f63cd95b219c56c8b703bcb10c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_factor_composed.txt
@@ -0,0 +1,6 @@
+Seja h = 14 - -8. Supõe h = -3*w + 91. Será 14 um fator de w?
+False
+Assume que n é 0/(-2)*2/4. Seja i = 9 - n. Seja x = i + 26. Será x um múltiplo de 17?
+False
+Seja m = 9 + -4. Supõe -m*q + 190 + 40 = 5*d, 0 = -4*d - 2*q + 194. Será d um múltiplo de 24?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1d4ca1f78dcb51f905e387325ff2c9197c0d8957
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime.txt
@@ -0,0 +1,6 @@
+Será 553 um número primo?
+False
+Será 139114 um número primo?
+False
+Será 21251 compósito?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..458f9e6ce04c56dae41aa133f6f42b1def77c297
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__is_prime_composed.txt
@@ -0,0 +1,6 @@
+Assume que n é 21/(2/4 - 2). Seja g(j) = -j - 9. Seja u, g(-6). Será (n/u)/(-2)*-3 um número primo?
+True
+Seja t = 9 + -15. Seja x(z) = z**2 - z - z + 9 + z + 6*z. Será x(t) um número compósito?
+True
+Será (-10820)/(-3) - 26/(-78) primo?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm.txt
new file mode 100644
index 0000000000000000000000000000000000000000..92b2d5398ec5d5d2653c1b5c78b1f1d749b8c3f4
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm.txt
@@ -0,0 +1,6 @@
+Calcula o denominador comum de 33/224 e 121/1120.
+1120
+Calcula o menor múltiplo comum de 1130 e 8.
+4520
+Calcula o menor múltiplo comum de 1792 e 512.
+3584
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..945eddc0738686ec65f61f758dbfe57b2fc8594f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__lcm_composed.txt
@@ -0,0 +1,6 @@
+Assume que w é 1/3*(9 + 0). Calcula o denominador comum de 77/18 e (-1)/3 - w/2.
+18
+Supõe -35 - 5 = -4*t. Supõe 15 = -4*q + j, 0*q = -4*q + 4*j - 12. Calcula o menor múltiplo comum de ((-2)/q)/((-4)/(-176)) e t.
+110
+Seja n(r) = 5 - 8*r. Seja q(u) = u - 1. Seja o(c) = -2*n(c) - 18*q(c). Seja t, o(10). Qual é o denominador comum de (17/20)/(t/(-30)) e 29/9?
+72
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9c774be667e4466bccbc806685bef00599ae270c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors.txt
@@ -0,0 +1,6 @@
+Quais são os fatores primos 235?
+5, 47
+Enumera os fatores primos de 675.
+3, 5
+Quais são os fatores primos 654?
+2, 3, 109
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f39a0df901cdcddb918e437b73cdbf7e0f7f9885
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__list_prime_factors_composed.txt
@@ -0,0 +1,6 @@
+Seja j(g) = g + 11. Seja q(d) = -d**2 - 5*d + 1. Supõe -4*r + 1 = -3*r, -2*l + 5*r = 17. Seja n, q(l). Quais são os fatores primos j(n)?
+2, 3
+Assume que y é -4 + 1 + 78/2. Supõe -4*i = -0*i - y. Enumera os fatores primos de i.
+3
+Assume que w é 7/14*(-4 + 0). Seja i(t) = 3*t**2 + 2*t - 2. Quais são os fatores primos i(w)?
+2, 3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d720c42af9c9f56a5fea035704e30e62d7f38f7e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value.txt
@@ -0,0 +1,6 @@
+Qual é o dígito das dezena milhares de 54091?
+5
+Qual é o dígito das milhares de 2360?
+2
+Qual é o dígito das dezenas de 5670?
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9f3e466e668ad4c24f70c33f0d98975d48b41412
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__place_value_composed.txt
@@ -0,0 +1,6 @@
+Qual é o dígito das unidades de 1815/(-66)*-8*(-2)/(-10)?
+4
+Seja u = -13 - -32. Qual é o dígito das unidades de u?
+9
+Seja b(x) = -x**2 + x + 9. Seja h(j) = 8 - j**2. Seja u(s) = 4*b(s) - 5*h(s). Qual é o dígito das unidades de u(-5)?
+1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2e3defa7b806c5efa300edd3b2bdf0ac481cda3e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number.txt
@@ -0,0 +1,6 @@
+Quanto é -0.00716 arredondado para as casa decimal das quatros?
+-0.0072
+Quanto é 510600 arredondado para as um centena milhar mais próxima?
+500000
+Arredonda -0.00595 para as casa decimal das quatros.
+-0.006
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..eeaf083272808d65353a315119fa3fec40caa341
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/numbers__round_number_composed.txt
@@ -0,0 +1,6 @@
+Supõe -8654 = -5*x - 4*a + 3*a, 2*x = a + 3456. Arredonda x para as 100 mais próxima.
+1700
+Supõe q = 3 + 126. Quanto é q arredondado para as dez mais próxima?
+130
+Seja x = 515 - 353. Seja p = -162.0091 + x. Arredonda p para as casa decimal das 3s.
+-0.009
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__add.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__add.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e555015086f2a664e728f98500bb25f1f2f0de98
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__add.txt
@@ -0,0 +1,6 @@
+Seja n(b) = -b**2 + b - 1. Seja u(t) = 2 + 15*t**2 + 0 - 7 + 6*t - 19*t**2. Supõe -w - 4*d - 14 = 0, 6 = -w - 2*d - 2. Determina w*u(q) + 10*n(q).
+-2*q**2 - 2*q
+Seja a(u) = -4*u - 6. Seja z, a(-3). Seja o(h) = 4*h**2 - 3. Seja y(i) = 7 - 9*i**2. Obtém z*y(g) + 14*o(g).
+2*g**2
+Seja f = -17 + 10. Seja o(s) = 5*s**3 + 6. Supõe -5*v + 0*n = -n - 20, v - 4*n = 4. Seja y(b) = -v + 11 + b**3 + 5*b**3. Determina f*o(c) + 6*y(c).
+c**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__coefficient_named.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__coefficient_named.txt
new file mode 100644
index 0000000000000000000000000000000000000000..df2a6da230fb3a58f09593a9c2c2925764045139
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__coefficient_named.txt
@@ -0,0 +1,6 @@
+Rearranja -o + 26*o**2 + 17*o**3 + 27*o**2 + 26*o**2 - 78*o**2 - 3 como j + y*o**2 + w*o + g*o**3 e dá y.
+1
+Rearranja -6*o**2 - 42*o**2 + 15*o**2 na forma x + p*o**2 + g*o e dá p.
+-33
+Expressa -1 + 41*n**2 - 3*n**4 + 0 - 2*n - 43*n**2 como j + y*n**4 + i*n**3 + d*n + t*n**2 e dá y.
+-3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__collect.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__collect.txt
new file mode 100644
index 0000000000000000000000000000000000000000..db737118caa48d237e0265de2ab96f0e4d6eb545
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__collect.txt
@@ -0,0 +1,6 @@
+Coleta os termos em 57*o - 36*o + 3*o**2 - 21*o.
+3*o**2
+Coleta os termos em 3*i + 35*i**3 - 29*i**3 - 9*i**3 - 3*i.
+-3*i**3
+Coleta os termos em -14*q**3 + 18*q**3 - 2*q**2 + 2*q**2.
+4*q**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__compose.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__compose.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d1c7305da642c590c9535a18edd84374dc4d566c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__compose.txt
@@ -0,0 +1,6 @@
+Seja n(a) = -7*a. Seja v(d) = -d**3 + 9*d**2 + d - 7. Seja w, v(9). Assume que p é (-4)/5*(-5)/2. Seja q(s) = -s**p + s**w - 2*s**2. O que é q(n(x))?
+-98*x**2
+Seja k(c) a segunda derivada de 3*c**4/4 - 4*c. Seja r(i) = -2*i**2 + 2*i + 2. Seja m(u) = u + 1. Seja s(h) = 2*m(h) - r(h). O que é s(k(n))?
+162*n**4
+Seja u(i) = i**2. Seja y(v) a terceira derivada de 0 + 1/20*v**5 + 0*v**4 + 7*v**2 + 0*v + 0*v**3. Determina u(y(j)).
+9*j**4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fce27889b9bbe36616b56c50ebee074a29a89d48
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate.txt
@@ -0,0 +1,6 @@
+Seja r(x) = -x**2 - 18*x - 15. Determina r(-18).
+-15
+Seja f(n) = -3*n**2 - 23*n - 1. Obtém f(-7).
+13
+Seja v(t) = 8 - 2*t**2. O que é v(6)?
+-64
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ac929730793ad07e365a10052f0ce4412e900b41
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__evaluate_composed.txt
@@ -0,0 +1,6 @@
+Assume que w é (-2)/7 + (-40)/7. Seja p(l) = -636*l + 159*l + 158*l + 164*l + 9 + 154*l. Determina p(w).
+15
+Seja l(h) = h**2 - 5*h + 4. Seja f(v) = 3*v**2 - 14*v + 13. Seja d(a) = 2*f(a) - 7*l(a). O que é d(6)?
+4
+Seja m(w) = -w**2 + w + 4. Seja q, m(0). Supõe -5*b = -u - 5, -2*b - 3*u + 3 = b. Seja o(i) = 2 + i - 3 - b. Determina o(q).
+2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__expand.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__expand.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f5e522382735c6538a2533d6468b68707ddd7262
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__expand.txt
@@ -0,0 +1,6 @@
+Expande (m - 1 + 1)*(-m**2 + 13*m**2 - 2*m**2 + (5*m - 5*m + 2*m)*(m - 2*m + 2*m)).
+12*m**3
+Expande ((2*x + 1 - 1)*(3*x - x - x) + 3*x**2 - x**2 - 3*x**2)*(-x - 2*x**2 + 6*x - 2*x).
+-2*x**4 + 3*x**3
+Expande (4*g**3 + 0*g**3 - 5*g**3)*(g - g - 2*g**2) + (12*g**3 + 16*g**3 - 6*g**3)*(0*g**2 - g**2 - g**2).
+-42*g**5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__simplify_power.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__simplify_power.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9e774df2b07f43d71e52fb58959006948bed8437
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/polynomials__simplify_power.txt
@@ -0,0 +1,6 @@
+Simplifica r*r**(-16)*r**(-1/7)*r assumindo que r é positiva.
+r**(-99/7)
+Simplifica ((k*k**7*k)/(k**(-1/5)*k))**(-1/4) assumindo que k é positiva.
+k**(-41/20)
+Simplifica (j/(j/(((j*j**(-2/23))/j)/j)))**(-3/4) assumindo que j é positiva.
+j**(75/92)
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_level_set.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_level_set.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5fac816956e67ab7f76b45a9499485e9b6d6905d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_level_set.txt
@@ -0,0 +1,6 @@
+Calcula da probabilidade de(a) escolher 2 k quando dois caractéres são escolhidos sem reposição de {k: 2, o: 3}.
+1/10
+Calcula da probabilidade de(a) escolher 4 h quando quatro caractéres são escolhidos sem reposição de {t: 7, h: 2}.
+0
+Calcula da probabilidade de(a) escolher 2 e e 1 d quando três caractéres são escolhidos sem reposição de eeeccctegccdceeect.
+7/272
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_sequence.txt b/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_sequence.txt
new file mode 100644
index 0000000000000000000000000000000000000000..974506f5fcf2f627f92bf3c337b96312623d198c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-easy/probability__swr_p_sequence.txt
@@ -0,0 +1,6 @@
+Qual é a probabilidade de(a) sequência hm quando dois caractéres são escolhidos sem reposição de {q: 1, h: 11, m: 7}?
+77/342
+Quatro caractéres são escolhidos sem reposição de alkkkslkkkeaaaahs. Qual é a probabilidade de(a) sequência eksa?
+1/952
+Quatro caractéres são escolhidos sem reposição de hnznjzq. Dá a probabilidade de(a) sequência nzhj.
+1/210
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0281162cdafbb92712aefcb3212c1441956f4d53
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d.txt
@@ -0,0 +1,6 @@
+Resolve -2532*x + 68646 - 953 = -8077 - 15382 para x.
+36
+Resolve -8075*a + 2436 = -8423*a para a.
+-7
+Resolve -2741574*o - 26705 = -2742119*o para o.
+49
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..090b210da868cdf5208999f0cd11770b8ad9e36c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_1d_composed.txt
@@ -0,0 +1,6 @@
+Seja f = -77 + 81. Seja o(z) = -z**3 + 13*z**2 - 22*z + 2. Seja q, o(11). Supõe -4*c + x + 4 = 0, -3*c + 4*x = -q*c - 16. Resolve -h + c*h + f = 0 para h.
+4
+Supõe 12*t = -6*t + 486. Seja o = -23 + t. Resolve m + 9 = o*m para m.
+3
+Seja o(d) = 543*d + 8. Seja q, o(1). Seja u = q - 543. Resolve 0 = u*v - 41 + 1 para v.
+5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..88e86f0c3f3e002698f2b54920053e5e884737de
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d.txt
@@ -0,0 +1,6 @@
+Resolve 0 = -4*w + 5*j - 12, w + 4*j = 5*w + 626 - 614 para w.
+-3
+Resolve -1076392*x + 1076395*x + 44 = -5*b, -3*b = 23*x + 514 para x.
+-23
+Resolve -11*n - 64*w + 30*w = -29*w - 165, 3*n + 3*w = 45 para n.
+15
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e809ed5854e9004aac9278a2dae0c2852af610d2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__linear_2d_composed.txt
@@ -0,0 +1,6 @@
+Supõe 0 = -17*c + 16*c + 16. Supõe v = -2*n + 11, -5*n + c + 8 = -v. Resolve 0 = 5*w - m - m - 3, 4*m = -w + n para w.
+1
+Supõe 5 = 5*o, -o = -2*d + 3*o + 6. Supõe -5*r = d*r - 3*r. Resolve r = w + 5*u - 3, 4 = -2*w + 2*u - 2 para w.
+-2
+Supõe 27*g - 3*g - 92 - 4 = 0. Resolve 22*z - 25*z + 5*n = 40, z - g*n + 25 = 0 para z.
+-5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots.txt
new file mode 100644
index 0000000000000000000000000000000000000000..071f9f1f561400efbe033c5280d6fc59275b2166
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots.txt
@@ -0,0 +1,6 @@
+Factoriza w**2/2 - 318351*w - 3183535/2.
+(w - 636707)*(w + 5)/2
+Factoriza -32*q**3 - 131363*q**2 - 143643*q - 12312.
+-(q + 1)*(q + 4104)*(32*q + 3)
+Factoriza c**3 - 5650*c**2 + 9789632*c - 4774711808.
+(c - 2416)**2*(c - 818)
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..dc27c88b8f6875b73805ade88d37c2234bc2b686
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__polynomial_roots_composed.txt
@@ -0,0 +1,6 @@
+Seja n = -920966/7 - -131570. Resolve 6/7*s**4 - 3/7*s**5 - 6/7 + 6/7*s**3 + 3*s - n*s**2 = 0 em relação a s.
+-2, 1
+Seja t(a) a segunda derivada de a**4/4 + 26*a**3 + 153*a**2/2 + 15*a - 1. Encontra w tal que t(w) = 0.
+-51, -1
+Supõe 0 = -2*q + 7*q + 4*l + 570, -25 = 5*l. Assume que s é (5/q)/(2/(-8)). Factoriza 0*n + 2/11 - s*n**2.
+-2*(n - 1)*(n + 1)/11
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_next_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_next_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fd49848d1d7a417ef085d4431752719d8810fa7e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_next_term.txt
@@ -0,0 +1,6 @@
+O que se segue em 96679, 192327, 287975?
+383623
+O que se segue em 67384555, 67384570, 67384593, 67384624, 67384663?
+67384710
+O que se segue em -854936, -854939, -854942?
+-854945
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_nth_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_nth_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e564632d61976acff24ba4152591cf433df8e29a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/algebra__sequence_nth_term.txt
@@ -0,0 +1,6 @@
+Qual é termo nº w na sequência 15613, 58800, 130777, 231544, 361101, 519448, 706585?
+14395*w**2 + 2*w + 1216
+Qual é termo nº q na sequência 92542, 185776, 279010, 372244, 465478, 558712?
+93234*q - 692
+Qual é termo nº c na sequência 5031, 20167, 45405, 80745, 126187, 181731, 247377?
+5051*c**2 - 17*c - 3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub.txt
new file mode 100644
index 0000000000000000000000000000000000000000..97e10586fe2c71650b26b0aae2fe091cbd799428
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub.txt
@@ -0,0 +1,6 @@
+Subtrai 0.14 de -3410.1745224.
+-3410.3145224
+Qual é a distância entre -729.9 e 55330904?
+55331633.9
+Calcula 1160326990177 - -178.
+1160326990355
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub_in_base.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub_in_base.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fa17606b91c001651d825783d709db899ed468f7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_or_sub_in_base.txt
@@ -0,0 +1,6 @@
+Em base 6, quanto é -10004225 - 4?
+-10004233
+Em base 16, quanto é 50eb + 7a4?
+588f
+Em base 15, quanto é -12c9 - 2045?
+-331e
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_sub_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_sub_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..996d9d617abb03edd0fda5d6ec57f6004427a5d6
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__add_sub_multiple.txt
@@ -0,0 +1,6 @@
+2 + 2 - (0 + (-9 - 7) + -5 + 145)
+-120
+Quanto é 23 - 0 - (-9 + (26 - -50))?
+-44
+Avalia -76 - (45 + -139) - (-22 - -2).
+38
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__div.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__div.txt
new file mode 100644
index 0000000000000000000000000000000000000000..54fe837b4d4b04986faedaac3869f68dc32d6d87
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__div.txt
@@ -0,0 +1,6 @@
+100734 dividido por -1
+-100734
+Divide -4 por 827002.
+-2/413501
+Divide -19108613 por 15.
+-19108613/15
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mixed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mixed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..31112670a60a75bc4d7e8fc641e3673874fcfd0a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mixed.txt
@@ -0,0 +1,6 @@
+Quanto é 16 + (31005/636)/((-342)/112)?
+2/57
+(-135)/(-2)*(45 - 1197/27)
+45
+(-355)/(-213)*(-2 + -7) + (-204)/(-4)
+36
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d84806ca5d25b95e03ca9bdc6f776e3e1b3036fa
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul.txt
@@ -0,0 +1,6 @@
+0.006985283*0.1
+0.0006985283
+Quanto é 0.7 vezes 385248?
+269673.6
+-264428119*-0.5
+132214059.5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul_div_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul_div_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2ab695779ff45fe66157c26d2d4fe94dce0b90a5
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__mul_div_multiple.txt
@@ -0,0 +1,6 @@
+Avalia (((-12627)/(-276))/3)/(2/(-8)).
+-61
+Qual é o valor de 10*(-2)/(-570)*((-1)/3)/(21/189)?
+-2/19
+Quanto é 1*60/(-11520)*-18?
+3/32
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__nearest_integer_root.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__nearest_integer_root.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2583a3b16a5810bfa74309428adbd4fa6f6ec603
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__nearest_integer_root.txt
@@ -0,0 +1,6 @@
+Quanto é 65443000 elevado a 1/7, arredondado para o inteiro mais próximo?
+13
+Qual é a terceira raíz de 146597612 arredondada para o inteiro mais próximo?
+527
+Qual é a quarta raíz de 2216814268 arredondada para o inteiro mais próximo?
+217
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__simplify_surd.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__simplify_surd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3ec96ad1c5c73e553db5f85187ed6b6190fd829e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/arithmetic__simplify_surd.txt
@@ -0,0 +1,6 @@
+Simplificar ((1*sqrt(3993)*-4 - ((-4*(sqrt(3993) - 1*sqrt(3993)) - sqrt(3993)) + sqrt(3993)))/(sqrt(12)*3 + sqrt(3) - (sqrt(3) + sqrt(108)/sqrt(4))))**2 + 3.
+21323/9
+Simplificar 1 + (sqrt(108) + -1)**2 + 5 - 4*sqrt(9)/(sqrt(21)/(sqrt(28)/sqrt(4))).
+115 - 16*sqrt(3)
+Simplificar (-3 + sqrt(176)*-5)**2 - (2 + sqrt(176) - sqrt(176))*-5 - ((3 + (2 + sqrt(176) - sqrt(176)) + sqrt(176)*-2*-5)**2 + 1).
+-13207 - 280*sqrt(11)
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7ed7aa60f4c8a472476ab6e927034786e10b213f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate.txt
@@ -0,0 +1,6 @@
+Qual é a terceira derivada de -5*b**2*u*w**3 - 11*b**2*u + 41*b**2*w**2 + 361*b + 117044*u*w**3 em relação a w?
+-30*b**2*u + 702264*u
+Qual é a terceira derivada de -26992083*c**4 - 340*c**2 + 2979*c - 2?
+-647809992*c
+Qual é a segunda derivada de -8361282*l**3 + 65031*l - 3?
+-50167692*l
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..697c876bbeb1e3122986fc7b232cd1a26dcb535c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/calculus__differentiate_composed.txt
@@ -0,0 +1,6 @@
+Seja b(u) = 2534*u**2 + 4009*u. Seja a(q) = -2528*q**2 - 4008*q. Seja f(l) = 5*a(l) + 4*b(l). Encontra a segunda derivada de f(k) em relação a k.
+-5008
+Supõe 3 - 4 = -4*h + 5*a, 3*h - 2*a = 6. Encontra a primeira derivada de -64 + 313*g**4 - 132*g**h - 120*g**4 em relação a g.
+244*g**3
+Supõe 14*d - 910 = -20300. Seja o = -900 - d. Qual é a derivada de 12*l + 15*l + 484 - o em relação a l?
+27
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7711d50565c2acc063defe0dde6d7105ba90f101
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest.txt
@@ -0,0 +1,6 @@
+Qual valor está mais perto de -27 em -1/6, 209, 0.05, -3?
+-3
+Qual está mais perto de -0.8?  (a) 161  (b) 0  (c) 4.4  (d) 1/20
+b
+Qual valor está mais perto de 0.27 em -0.18, 2, 163, -0.1, 0.2?
+0.2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ef42e39c8be5f1a2b5e285ad48b59bc826c95ca0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__closest_composed.txt
@@ -0,0 +1,6 @@
+Seja g = 43.109 - 0.109. Seja x = g - 47. Seja v = 0.74 - 0.24. Qual valor está mais perto de -0.1 em v, -0.01, x?
+-0.01
+Supõe 0*d = 3*d - 147. Seja y = -244/5 + d. Qual valor está mais próximo de -1 em -0.1, y, -8?
+-0.1
+Supõe 6*i = 3*i - 6. Assume que g é ((-5)/(-4))/(i/(-8)). Assume que u é 4 + (100/(-30))/1. Qual está mais próximo de u?  (a) 4  (b) 0.4  (c) g
+b
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..202f300ac00971dbf387450f193bc01edc4b2388
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest.txt
@@ -0,0 +1,6 @@
+Qual é o quarto menor valor em 0.9, -5, -403, 5, -47?
+0.9
+Qual é o maior valor em -7, 0.034, 1542.2?
+1542.2
+Qual valor é terceiro maior?  (a) -0.04  (b) 1380  (c) -19  (d) -0.042  (e) 1
+a
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9210a17bf106fa9fc3671021c59b428980202c63
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__kth_biggest_composed.txt
@@ -0,0 +1,6 @@
+Seja p = -3815 - -3815.37. Qual é o terceiro maior valor em -1/7, -0.32, p?
+-0.32
+Seja j = 409927/39 - 10511. Qual é o menor valor em j, -8/25, 5, -3?
+-3
+Seja t = 1.88 - 5.56. Seja p = t + -0.32. Seja w = 0.04 - 2.04. Qual é o maior valor em w, 16, p?
+16
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ce16ff3e4502a010c1fff7f22159e0459cd89631
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair.txt
@@ -0,0 +1,6 @@
+Será que 3521/18908 e -1 possuem o mesmo valor?
+False
+Qual é maior: -1 ou -9/4030535?
+-9/4030535
+Será -0.02 <= -33887?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6e2a47ce9b10deb1a9b560dbf93c444f1103438d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__pair_composed.txt
@@ -0,0 +1,6 @@
+Assume que y é (-4)/(-10) - (-52)/(-5). Seja k = -317.834 + 329.834. Será y menor do que k?
+True
+Supõe 4*w + 5*c + 115 = -193, 236 = -3*w - 5*c. Qual é superior: -68 ou w?
+-68
+Seja v(q) = 2074*q + 10370. Seja w, v(-5). Qual é superior: 56/321 ou w?
+56/321
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1f4fbe5eb16a1924e300386c85edebde35143d40
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort.txt
@@ -0,0 +1,6 @@
+Coloca -479, 3, 19, -1 em ordem crescente.
+-479, -1, 3, 19
+Ordena 4023, 5, 105, -8, -3 por ordem decrescente.
+4023, 105, 5, -3, -8
+Coloca -89579, -86, -7 em ordem descendente.
+-7, -86, -89579
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e8f3f30ed46cdf26dafed3b2c83a419568adbda7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/comparison__sort_composed.txt
@@ -0,0 +1,6 @@
+Seja a(z) = z**2 - 5*z - 6. Seja w, a(5). Seja g(k) = 31*k + 4*k + 2*k**2 - 4*k + 10. Seja c, g(-15). Coloca -4, c, w em ordem decrescente.
+-4, c, w
+Seja r(o) = o**3 - o**2 + o. Seja i, r(0). Supõe -3*b + 21 = d, 43 = -3042*b + 3046*b + 3*d. Ordena b, -11, i por ordem decrescente.
+b, i, -11
+Seja d = 4.43 - 4.13. Seja q = 16.3 - d. Ordena q, -1, -6 por ordem decrescente.
+q, -1, -6
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0b2649d7af33adc70eaa0d69b936c1c50ae32815
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__conversion.txt
@@ -0,0 +1,6 @@
+Quanto é 42/5 de um(a) litro em um(a) mililitros?
+8400
+Quanto é 0.2856971 dias em nanosegundos?
+24684229440000
+Quanto é 23/8 de um(a) semana em um(a) minutos?
+28980
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__time.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__time.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0f761bef281f71685f839c7cf88d87869d8926fc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/measurement__time.txt
@@ -0,0 +1,6 @@
+Quantos minutos existem entre 7:00 AM e 7:35 AM?
+35
+O que é 120 minutos antes de 12:40 AM?
+10:40 PM
+Quantos minutos existem entre 4:23 AM e 3:07 PM?
+644
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__base_conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__base_conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5b691ca4149f16da00f6784ac86f887123a0c204
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__base_conversion.txt
@@ -0,0 +1,6 @@
+Quanto é 112021211211022 (base 3) em base 11?
+4317a28
+Quanto é -10051126 (base 8) em base 14?
+-3d1d42
+115fe2 (em base 16) para base 8
+4257742
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0a7db9f38c6921efc129b8a9fd47b294efd42d22
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder.txt
@@ -0,0 +1,6 @@
+Calcula o resto da divisão entre 7067451 e 23.
+11
+Calcula o resto da divisão entre 167204 e 40.
+4
+Calcula o resto da divisão entre 348056 e 5705.
+51
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ee8d9c2960cc723f88b11ac3fbd3520f7eb2b618
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__div_remainder_composed.txt
@@ -0,0 +1,6 @@
+Qual é o resto da divisão entre -6 + (1667 + 58 - (-3 - (1 + 3))) e 12?
+10
+Supõe 181*s - 179*s = 198. Supõe -180 = -9*z + s. Qual é o resto da divisão entre z e 18?
+13
+Seja y(v) = v**3 - v - 89. Seja o, y(0). Seja j = o - -59. Calcula o resto da divisão entre 16 e j/4*(-16)/20.
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9e1ba4fb2e0a1dbddc1037afbcc277ac48b0e186
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd.txt
@@ -0,0 +1,6 @@
+Qual é o maior divisor comum  de 3354966 e 972?
+54
+Calcula o maior divisor comum  de 206 e 4250501.
+103
+Calcula o maior fator comum  de 64863 e 7569.
+9
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e92ae26d007638d402c4ae68baa66d76a7842838
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__gcd_composed.txt
@@ -0,0 +1,6 @@
+Seja d(y) a segunda derivada de y**5/20 - y**4/3 - y**3/6 + 4*y**2 - 7*y. Seja p, d(4). Supõe h + 18 = p*h. Qual é o maior divisor comum  de h e 54?
+6
+Seja s(p) = p**3 - 4*p**2 - 2*p - 2. Seja a, s(5). Seja o = 167 + -110. Seja b = o - 5. Calcula o maior fator comum  de b e a.
+13
+Seja t(x) = -x**2 + 24*x - 18. Seja o, t(32). Seja g = 277 + o. Qual é o maior fator comum  de g e 5?
+1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor.txt
new file mode 100644
index 0000000000000000000000000000000000000000..50e9d42470c627b3c111e3cdcfedd00bf6212b87
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor.txt
@@ -0,0 +1,6 @@
+Será que 1897 divide 129639083?
+True
+Será 3388 um fator de 68522300?
+True
+Será 2274990355 um múltiplo de 151?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..196ecb3c6895904ca65a9c8ee744e6e8e8a44f32
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_factor_composed.txt
@@ -0,0 +1,6 @@
+Assume que r é (-3)/15*(-5)/(-2)*8. Será 31 um fator de 4*(6/r + (-8 - -66))?
+False
+Seja u(w) = w**2 - 4*w - 29. Seja v, u(-4). Supõe -3*d = -v, 2*m - 138 = -d + 3*d. Será 17 um fator de m?
+False
+Supõe -374*c - 358*c + 719*c + 221208 = 0. Será que 12 divide c?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime.txt
new file mode 100644
index 0000000000000000000000000000000000000000..83649f45ffaee44d1c6236fc4d7c98cd17616a87
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime.txt
@@ -0,0 +1,6 @@
+Será 15164814809 um número primo?
+False
+Será 50244651731 um número primo?
+False
+Será 80964269 compósito?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..00828be3ef27cd7b39e410c78953a6a24094b439
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__is_prime_composed.txt
@@ -0,0 +1,6 @@
+Assume que z é (0 + (-6)/4)/(21/(-70)). Seja x = -3 + z. Supõe x*q - 534 = 1984. Será q um número primo?
+True
+Supõe 3*w + 15 = 2*x, -69*x - 5*w - 29 = -71*x. Assume que q é 1*-15*-9*41. Será (q - -6)*(-1)/x primo?
+True
+Supõe -2*p + 1165237 = -848457. Será p um número primo?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2cd2c5ca48bc1a6a70f32fe43d0d31e08e0282c9
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm.txt
@@ -0,0 +1,6 @@
+Encontra o denominador comum de -26/2089055227 e -46/9.
+18801497043
+Calcula o denominador comum de -77/15053842 e -13/11939254.
+346238366
+Calcula o menor múltiplo comum de 1947328 e 7464.
+1816857024
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c3046460a52d23a980ee9556998b1a514495d334
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__lcm_composed.txt
@@ -0,0 +1,6 @@
+Seja b(v) = -v - 24. Seja a(k) = -2*k - 24. Seja f(z) = 3*a(z) - 4*b(z). Supõe -3*u - 5*i + 12 = 0, u - 8 + 4 = 4*i. Qual é o menor múltiplo comum de f(8) e u?
+8
+Seja i(n) a primeira derivada de -n**4/4 + 7*n**3/3 + n**2 + 4*n - 1. Seja u(s) = -s**2 - 93*s - 426. Qual é o menor múltiplo comum de u(-88) e i(7)?
+126
+Assume que a é (2/1 + -3)*(-2 - -2). Assume que l é 2 + a - 2 - -2. Encontra o denominador comum de -46/15 e (l - 2) + -1 + (-66)/45.
+15
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f6dad0ba68f0b48e200a7fc3730d8cf63c1b6192
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors.txt
@@ -0,0 +1,6 @@
+Quais são os fatores primos 209225851?
+17, 31, 397013
+Quais são os fatores primos 1261838204?
+2, 11, 53, 541097
+Enumera os fatores primos de 236683397.
+59, 4011583
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fbf8b5dc663bcddec4879591fff89b91f65af05d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__list_prime_factors_composed.txt
@@ -0,0 +1,6 @@
+Supõe -17*x = 5 + 29. Seja t(y) = -3*y**2 - 3*y. Seja j, t(x). Seja z(g) = 2 - 38*g. Quais são os fatores primos z(j)?
+2, 5, 23
+Assume que a é -8*((-2)/4 - -1). Seja i(t) a terceira derivada de -13*t**4/24 - 3*t**3/2 - 12*t**2 - 5. Quais são os fatores primos i(a)?
+43
+Assume que a é (6 - 7)/(3 - 10/3). Assume que y é 840/72 - (8/a - 2). Supõe -y*u + 7*u = -876. Quais são os fatores primos u?
+3, 73
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value.txt
new file mode 100644
index 0000000000000000000000000000000000000000..190434925022fb81c580581de7a729934514939f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value.txt
@@ -0,0 +1,6 @@
+Qual é o dígito das dezenas de 1862597320?
+2
+Qual é o dígito das milhares de 2210669121?
+9
+Qual é o dígito das centenas de 142920283?
+2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9dc0567c36c62f9acdcca3cccce65716152185fe
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__place_value_composed.txt
@@ -0,0 +1,6 @@
+Supõe -5*m - d + 3 = -14, 10 = m - 2*d. Seja c = 4 - 1. Supõe 0*b + 78 = 5*z - m*b, 2*b - 38 = -c*z. Qual é o dígito das unidades de z?
+4
+Seja i = -21495 - -26319. Qual é o dígito das milhares de i?
+4
+Supõe -8*b = h - 5*b - 3, 5*h - 5*b = -5. Supõe -4*w + 113 + 2047 = h. Qual é o dígito das dezenas de w?
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number.txt
new file mode 100644
index 0000000000000000000000000000000000000000..fe435d67c3b35df920ea84d5ca3289cc28813f6e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number.txt
@@ -0,0 +1,6 @@
+Quanto é 23494844.88 arredondado para as 10000 mais próxima?
+23490000
+Arredonda 460.9310374 para as dez mais próxima.
+460
+Arredonda 428.44828 para as um centena mais próxima.
+400
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1c3f5d93b63c12265b32b991bfb9bdce8e64b144
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/numbers__round_number_composed.txt
@@ -0,0 +1,6 @@
+Seja r = -2416 - -2409.501. Quanto é r arredondado para as inteiro mais próximo?
+-6
+Supõe 32*p - 34*p - 10753 = -5*a, 4*p = -4*a + 8580. Arredonda a para as um centena mais próxima.
+2100
+Supõe 0 = -3*v - 5*w - 14, 3*v = -10*w + 8*w - 2. Supõe z - 1805004 = v*b, -26*z + 31*z - 4*b - 9025008 = 0. Quanto é z arredondado para as 100000 mais próxima?
+1800000
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__add.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__add.txt
new file mode 100644
index 0000000000000000000000000000000000000000..91fbd1a2a06f3f95eb1f2778f643fac6e07b0b93
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__add.txt
@@ -0,0 +1,6 @@
+Seja f(x) = 14*x - 1793. Seja a(k) = 3*k - 2. Calcula 5*a(i) - f(i).
+i + 1783
+Seja w = -3635 - -3631. Seja u(b) = -b**3 - 4*b**2 + 9. Seja t(i) = 2*i**3 + 9*i**2 - 19. Assume que g é 1*2/(6/(-27)). Determina g*u(j) + w*t(j).
+j**3 - 5
+Assume que r é (-1830)/57 + 33 + (2 - 36/19). Seja p(l) = 1. Seja b(x) = 10*x + 2. Determina r*p(w) - b(w).
+-10*w - 1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__coefficient_named.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__coefficient_named.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6df40148f6c8124906aa34ea2e29b606ea3852c8
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__coefficient_named.txt
@@ -0,0 +1,6 @@
+Expressa (3*a**3 - 4*a**3 - 2*a**3)*(1 + 3 - 2) + 3*a**3 - 13*a**3 + 3*a**3 + (4 + 0 + 1)*(0*a + 0*a + a**3) como q*a**3 + m*a + c + v*a**2 e dá q.
+-8
+Expressa 598*d + 15*d**2 + 1 - 299*d + 0 - 4 - 280*d na forma j + z*d**2 + b*d e dá z.
+15
+Rearranja (11*a - 13*a + 15 + 0)*(1 + 6 + 0)*(0 + 0 - 3)*(6 + 0 + 8) na forma b*a + d e dá b.
+588
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__collect.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__collect.txt
new file mode 100644
index 0000000000000000000000000000000000000000..142a24cba62d040b692c5cd9b35a94b1f3eedbaf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__collect.txt
@@ -0,0 +1,6 @@
+Coleta os termos em -12 - 51 + 63 - 268*c - 66*c + 21*c.
+-313*c
+Coleta os termos em 17*i - 36*i + 24*i + 133*i + 9*i.
+147*i
+Coleta os termos em 301 - 301 + 64316*q - 64314*q + 18*q**3 - q**2 + q**2.
+18*q**3 + 2*q
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__compose.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__compose.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5937abf9ddb47ed015f6e8b61eb1144c00f7bb9d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__compose.txt
@@ -0,0 +1,6 @@
+Seja n(b) = 5 - 12*b. Seja a(u) a primeira derivada de 4*u**3 - 1019. Calcula n(a(w)).
+5 - 144*w**2
+Seja d(p) = -p**3 - 11*p**2 - p - 3. Seja n, d(-11). Seja s(x) = -15 - x + 5 - n - 3. Seja t(q) = -3*q. Determina s(t(o)).
+3*o - 21
+Seja g(m) a primeira derivada de -4*m**2 - 10*m + 9495. Seja k(x) = 97*x. O que é g(k(t))?
+-776*t - 10
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..857a213bf76aec0610fd7b12549bcae8ed97df6b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate.txt
@@ -0,0 +1,6 @@
+Seja p(f) = 593*f**2 - 8350*f + 14. Obtém p(14).
+-658
+Seja t(x) = 30893*x - 2471431. Calcula t(80).
+9
+Seja g(c) = 356*c + 11775. O que é g(-33)?
+27
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c3887d90ce7d921d3359800147f8ea5a3e28444f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__evaluate_composed.txt
@@ -0,0 +1,6 @@
+Supõe q - 22 - 32 = 0. Seja d(h) = -4*h + 126 - 65 - q + h**2 - h. O que é d(7)?
+21
+Seja g(z) = -z**3 + 3*z**2 + 2*z + 1. Supõe -102*k = -107*k + 30. Supõe -k*q + 168 = q. Supõe 5*l + 3*l - q = 0. Obtém g(l).
+7
+Seja t(f) a segunda derivada de -7*f + 1 + 1/20*f**5 + 19/6*f**3 - 11/2*f**2 - 5/6*f**4. Calcula t(8).
+13
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__expand.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__expand.txt
new file mode 100644
index 0000000000000000000000000000000000000000..0101f12b3d041f48f18097c4d9e01b7e1a6ea15e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__expand.txt
@@ -0,0 +1,6 @@
+Expande (4460*i + 1780*i - 7497*i + 4778*i)*(6 + 2 + 5).
+45773*i
+Expande 287*p + 1382601 - 1382601 + (0 - 2 + 1)*(-p + 1 - 1) + 2*p + p - 2*p + p - 2*p + 2*p.
+290*p
+Expande (-228 - 164 - 158 + 30)*(u**2 - 4*u**3 - 2*u**3 + 12*u**3).
+-3120*u**3 - 520*u**2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__simplify_power.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__simplify_power.txt
new file mode 100644
index 0000000000000000000000000000000000000000..82632a00f374a754764ca873bc1a413dd16ec7ce
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/polynomials__simplify_power.txt
@@ -0,0 +1,6 @@
+Simplifica ((v/v**3*v)**(2/27))**18*(v**(-1))**44*v/(v/((v*(v/(v*v*v**(-7)/v*v*v))/v*v)/v*v))*v*v**(2/19)*v assumindo que v é positiva.
+v**(-2122/57)
+Simplifica ((n/n**(-4/11))/(n/n**(1/18))*(n*n**11)**(-2/37))**(-30/13) assumindo que n é positiva.
+n**(8405/15873)
+Simplifica ((n/n**(-1)*n)/(n*n**(6/17))*n**10*n*n*n**(-3/4))**(-10) assumindo que n é positiva.
+n**(-4385/34)
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_level_set.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_level_set.txt
new file mode 100644
index 0000000000000000000000000000000000000000..28f27d12792b1e3f740875a51d5c2679075cc732
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_level_set.txt
@@ -0,0 +1,6 @@
+Dois caractéres são escolhidos sem reposição de djpp. Dá a probabilidade de(a) escolher 1 p e 1 j.
+1/3
+Calcula da probabilidade de(a) escolher 1 f e 3 i quando quatro caractéres são escolhidos sem reposição de fyifyizyiyiyeeyye.
+2/595
+Qual é a probabilidade de(a) escolher 1 k, 1 c e 1 j quando três caractéres são escolhidos sem reposição de ckecjnne?
+1/28
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_sequence.txt b/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_sequence.txt
new file mode 100644
index 0000000000000000000000000000000000000000..881d98e7c71fc62e3f8697ca23efdd39108367b1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-hard/probability__swr_p_sequence.txt
@@ -0,0 +1,6 @@
+Quatro caractéres são escolhidos sem reposição de {r: 1, d: 4, y: 1, l: 1, f: 2}. Qual é a probabilidade de(a) sequência fydf?
+1/378
+Três caractéres são escolhidos sem reposição de bbyyvyybv. Qual é a probabilidade de(a) sequência vbb?
+1/42
+Calcula da probabilidade de(a) sequência wnk quando três caractéres são escolhidos sem reposição de {w: 2, f: 1, k: 2, n: 1}.
+1/30
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f4629f3a26df9d38ce0709fac8e20b1d98590da1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d.txt
@@ -0,0 +1,6 @@
+Resolve -26*s + 6402 = 121*s + 144*s para s.
+22
+Resolve 42959*d - 42919*d = -720 para d.
+-18
+Resolve 278 + 539 - 47 = 385*d para d.
+2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..cbf4aa24cb1dd7618f94ac658877340a26139a06
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_1d_composed.txt
@@ -0,0 +1,6 @@
+Seja o = 15 + -10. Supõe -9 = 3*f - 15. Supõe 11 = 5*h + 2*p, -f*h + 2*p + p = -12. Resolve o = -h*i + 2 para i.
+-1
+Seja s = 792 + -785. Supõe 0 = j + j - 56. Assume que m é s*8/j*1. Resolve 0 = m*y + 3 + 5 para y.
+-4
+Seja h(y) = -y**3 + y**2 + y + 4. Seja t, h(0). Supõe -30 = -2*p + 5*f, p + 4*f = -3*p + t. Assume que x é p*(-3 - (3 - 7)). Resolve -q - 12 = -x*q para q.
+3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d.txt
new file mode 100644
index 0000000000000000000000000000000000000000..2cca63cb181673e0a89b8122a1af921da8ac4f7f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d.txt
@@ -0,0 +1,6 @@
+Resolve -2*z + 7 = 3*h, -4*z = -2360*h + 2364*h - 4 para h.
+5
+Resolve -5*r + 5 = -48*p + 50*p, -r + p = -1 para r.
+1
+Resolve i + 5*i = 5*i - 5*l - 47, 0 = l + 9 para i.
+-2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6e21c35b7d8ae40b787aa7445990be0b5966710d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__linear_2d_composed.txt
@@ -0,0 +1,6 @@
+Seja g(s) = 4*s**2 - 21*s - 5. Seja m, g(5). Assume que h é 4/m + 18/45. Resolve -5*r - 5 = -3*c, h = 3*c - 6*c + r - 11 para c.
+-5
+Seja u(s) = 7 - s. Supõe -35 = 5*z + 25. Seja l, u(z). Resolve -5*j - 12 = 4*w - 3, w = 3*j + l para j.
+-5
+Assume que a é 1/(2/((-44)/(-2))). Supõe -5*y = -3*b - y - a, -10 = -2*y. Supõe n = b*n - 32. Resolve 5*w - 6 - 9 = 0, -u = 4*w - n para u.
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7b5ece38a51724ea65b4378cec2a0142b69e1239
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots.txt
@@ -0,0 +1,6 @@
+Determina l, dado que 25*l**4 + 7840*l**3 + 613086*l**2 - 246176*l + 24649 = 0.
+-157, 1/5
+Factoriza -3*q**5/7 + 96*q**4/7 - 183*q**3/7 + 90*q**2/7.
+-3*q**2*(q - 30)*(q - 1)**2/7
+Determina a, dado que 3*a**2/5 - 1773*a/5 + 3534/5 = 0.
+2, 589
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..deba19dca66fed3cfb75027ad8e069a23a02ccef
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__polynomial_roots_composed.txt
@@ -0,0 +1,6 @@
+Seja m(p) a primeira derivada de 17 + 0*p + 3*p**5 + 5/3*p**3 + 0*p**2 - 5*p**4. Seja m(c) = 0. Qual é c?
+0, 1/3, 1
+Seja l(q) a primeira derivada de 1/10*q**5 + q**2 - 1/2*q**3 - 9 - 1/4*q**4 + 2*q. Factoriza l(i).
+(i - 2)**2*(i + 1)**2/2
+Assume que x é 0 - ((-2)/1 - -8). Seja g(n) = -n - 4. Seja d, g(x). Seja -2 + 1 - 5*b**2 + b**3 - b + 6*b**d = 0. Qual é b?
+-1, 1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_next_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_next_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..392e9313db0ee5ef212f4eac1f02e886d1e4f907
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_next_term.txt
@@ -0,0 +1,6 @@
+O que se segue em 15, -21, -69, -129?
+-201
+O que se segue em 1644, 3286, 4926, 6564, 8200, 9834, 11466?
+13096
+O que se segue em 38808, 38807, 38806, 38805, 38804, 38803?
+38802
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_nth_term.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_nth_term.txt
new file mode 100644
index 0000000000000000000000000000000000000000..93b4f3a72a7d36aeb542d65e1ed31fd06610e47d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/algebra__sequence_nth_term.txt
@@ -0,0 +1,6 @@
+Qual é termo nº y na sequência -4397, -8806, -13223, -17648?
+-4*y**2 - 4397*y + 4
+Qual é termo nº r na sequência -2432, -2418, -2404, -2390, -2376, -2362?
+14*r - 2446
+Qual é termo nº n na sequência 533, 1025, 1425, 1685, 1757, 1593, 1145?
+-8*n**3 + 2*n**2 + 542*n - 3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1ff6571f2040c829febe494c247066bf60ecd073
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub.txt
@@ -0,0 +1,6 @@
+Qual é a distância entre -2855 e -2.47099?
+2852.52901
+Quanto é -6.1414 - 0.051822?
+-6.193222
+-73+-12559367.6
+-12559440.6
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub_in_base.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub_in_base.txt
new file mode 100644
index 0000000000000000000000000000000000000000..09c9cbd7e51c693e6f16bb57d3c2c7c6f7ea171b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_or_sub_in_base.txt
@@ -0,0 +1,6 @@
+Em base 15, quanto é -1 - 45e7?
+-45e8
+Em base 5, quanto é 22 - -22414?
+22441
+Em base 16, quanto é 0 - -2074?
+2074
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_sub_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_sub_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ca0d2dce2f6d591323f01c7baea38895feaad4d7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__add_sub_multiple.txt
@@ -0,0 +1,6 @@
+Avalia -6 + (-11 - -17) - 19.
+-19
+Avalia -3 - (26 + -9 - 8) - 2.
+-14
+Calcula -3 + (14 - 10 - (2 + 15 + 0)).
+-16
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__div.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__div.txt
new file mode 100644
index 0000000000000000000000000000000000000000..de57bc42a33f9e271b9c2366fa5b9873349deebc
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__div.txt
@@ -0,0 +1,6 @@
+Quanto é -84132 dividido por 114?
+-738
+Divide -904 por -808.
+113/101
+Divide -52606 por -26303.
+2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mixed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mixed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..27e69ba40404b928bfe82011038841ce652c23d2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mixed.txt
@@ -0,0 +1,6 @@
+Avalia (5/(-15)*(0 + 15))/7.
+-5/7
+Quanto é (-3515)/323 - (-6)/(-51)?
+-11
+Avalia ((-426)/(-90))/((-6)/10) - (-12)/54.
+-23/3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a7242ca39b3be08fb59ef19a34c71fb575a0bf35
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul.txt
@@ -0,0 +1,6 @@
+264.5 * 7
+1851.5
+-0.5 vezes -708.9
+354.45
+853813 vezes -0.2
+-170762.6
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul_div_multiple.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul_div_multiple.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9a40a555cdc98220516f08c11374619530dd1388
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__mul_div_multiple.txt
@@ -0,0 +1,6 @@
+Quanto é (207/3933)/((-1)/(-3))?
+3/19
+Calcula (28*(-33)/(-99))/((-4)/15).
+-35
+Qual é o valor de ((-5)/(-7))/(3/3)?
+5/7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__nearest_integer_root.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__nearest_integer_root.txt
new file mode 100644
index 0000000000000000000000000000000000000000..3f3fac19caa8c112a46376198834d8b71924c72d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__nearest_integer_root.txt
@@ -0,0 +1,6 @@
+Qual é a quinta raíz de 805081 arredondada para o inteiro mais próximo?
+15
+Quanto é 6819931 elevado a 1/3, arredondado para o inteiro mais próximo?
+190
+Qual é a raíz quadrada de 451226 arredondada para o inteiro mais próximo?
+672
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__simplify_surd.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__simplify_surd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ff1af8c1f2b769bb221e2840324c446ec827020b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/arithmetic__simplify_surd.txt
@@ -0,0 +1,6 @@
+Simplificar ((sqrt(196) - (sqrt(196)*-4 - sqrt(196)))/sqrt(4))/(sqrt(448)*-4).
+-3*sqrt(7)/16
+Simplificar (-4 + 0 + sqrt(5) + 5 - (2*sqrt(120))/sqrt(6))**2.
+46 - 6*sqrt(5)
+Simplificar (-1*sqrt(180)*-1 - (2*sqrt(180)*3 - sqrt(180)))**2.
+2880
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..5e201add61fedd37bf74f108fc87040660328697
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate.txt
@@ -0,0 +1,6 @@
+Qual é a primeira derivada de 3310*d*z**2 - 218*d*z + 101*z**2 - 1188 em relação a d?
+3310*z**2 - 218*z
+Qual é a segunda derivada de -s**3*w + 3060*s**3 + 2*s**2*w - 3*s*w - 60036*w em relação a s?
+-6*s*w + 18360*s + 4*w
+Deriva -a*h**3*t - 460274*a*h + h**3 + 6*h**2*t - 2*h*t + h - 13335*t em relação a a.
+-h**3*t - 460274*h
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..14c41bb5b82ab2fb34c8dd61692a4e6ed4dd86a9
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/calculus__differentiate_composed.txt
@@ -0,0 +1,6 @@
+Diferencia -23*m**4 - 64*m**4 - 4*m**4 - 129 + 6*m**4 em relação a m.
+-340*m**3
+Seja l(z) a terceira derivada de -73*z**7/42 - z**5/60 - z**4/6 - 308*z**2 - 2. Encontra a terceira derivada de l(f) em relação a f.
+-8760*f
+Qual é a segunda derivada de 2 - 111*y**5 - 62*y**5 - 158*y**5 - 371*y - 94*y**5 em relação a y?
+-8500*y**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..d4b43486c245a0e27e8bf4b4154a0f5a318be74d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest.txt
@@ -0,0 +1,6 @@
+Qual está mais perto de -75?  (a) 9  (b) -3  (c) -0.1038
+b
+Qual valor está mais próximo de -44 em -2/9, -1/3, -3?
+-3
+Qual valor está mais perto de 0 em -5/2, -1, -1/2, 0?
+0
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4236c5e40eb469a53a3c38126a7c612b86bd0f1c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__closest_composed.txt
@@ -0,0 +1,6 @@
+Supõe 0 = -45*a - 5 + 5. Qual valor está mais perto de -1 em -0.1, -1, a?
+-1
+Assume que s é ((-9)/24)/((-1)/2). Seja q = 61 - 425/7. Seja x = 4 - 0. Qual valor está mais próximo de q em x, s, 3?
+s
+Seja y = 10.9 - 10.4. Qual valor está mais próximo de 0 em y, -1, -2?
+y
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest.txt
new file mode 100644
index 0000000000000000000000000000000000000000..aa844f8305977688f29a374b147a267f0fa3c47d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest.txt
@@ -0,0 +1,6 @@
+Qual valor é segundo maior?  (a) -17  (b) -1/6  (c) -103  (d) 0.3
+b
+Qual é o quinto maior valor em -0.03, -4, 0.4, -2/9585, -0.3?
+-4
+Qual valor é maior?  (a) 14/79  (b) 4  (c) -4  (d) -8
+b
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..37d883d9ab68014558226405e2c8f96114928713
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__kth_biggest_composed.txt
@@ -0,0 +1,6 @@
+Assume que z é ((-3)/(-63))/(3/9). Qual é o segundo maior valor em z, 2/7, 2/3, 0.32?
+0.32
+Seja m = 220.057 + -0.057. Qual é o quarto maior valor em -2, m, -0.2, -5?
+-5
+Supõe 264 = -4*v + 4*y, 4*y - 54 = v + 6*y. Seja q = -57 - v. Qual valor é segundo menor?  (a) 19  (b) -3  (c) q  (d) -5
+b
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair.txt
new file mode 100644
index 0000000000000000000000000000000000000000..9fa69d112bfc6485379606ff90e9179875a8609d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair.txt
@@ -0,0 +1,6 @@
+Qual é maior: 10030 ou 10016?
+10030
+Será -178, no máximo, -241?
+False
+Será 8 >= -10545?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..85f5c9404334044a2962a842033e9594eec61a69
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__pair_composed.txt
@@ -0,0 +1,6 @@
+Assume que n é 1/(-36)*252/1134. Qual é menor: n ou 0?
+n
+Seja y = 1545645/52 - 117885/4. Seja r = y - 253. Supõe -4*m - 8 + 4 = 0. Qual é superior: m ou r?
+r
+Assume que p é 12/3 + (-18 - (-28)/4). Qual é superior: p ou 15?
+15
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1f8d42f881fbdabd19e6a1454342cd905ae858e6
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort.txt
@@ -0,0 +1,6 @@
+Ordena -6, 9, 11067 por ordem crescente.
+-6, 9, 11067
+Coloca 2, 0.34, -1, -1/6, 3 em ordem decrescente.
+3, 2, 0.34, -1/6, -1
+Ordena -73, 3/5, -5, -1.4 por ordem ascendente.
+-73, -5, -1.4, 3/5
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..369ead2bf71ca76043e8e12108d196da5aecce02
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/comparison__sort_composed.txt
@@ -0,0 +1,6 @@
+Seja u = 443 - 446. Ordena 0.4, u, -0.059, -0.2 por ordem decrescente.
+0.4, -0.059, -0.2, u
+Seja p = -20 - 8. Seja b = -28.3 - p. Ordena b, -0.5, 0.1 por ordem decrescente.
+0.1, b, -0.5
+Seja l = 12.64 + -22.64. Seja j = -1 - -1. Ordena l, 0.2, j.
+l, j, 0.2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..83a20e522b205142462ec59fba32215edcdf25ca
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__conversion.txt
@@ -0,0 +1,6 @@
+Quantos(as) mililitros estão em 3/10 de um(a) litro?
+300
+Converte 2.0196201 meses para anos.
+0.168301675
+Quanto é 24705.99mm em centímetros?
+2470.599
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__time.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__time.txt
new file mode 100644
index 0000000000000000000000000000000000000000..06d48076cffd73d2c5174afebf022544f9804d74
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/measurement__time.txt
@@ -0,0 +1,6 @@
+O que é 589 minutos antes de 6:54 PM?
+9:05 AM
+Quantos minutos existem entre 10:19 AM e 6:41 PM?
+502
+Quantos minutos existem entre 10:07 PM e 4:01 AM?
+354
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__base_conversion.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__base_conversion.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b2bd384a9faa722cf8b4c8f1cec0de5f30dd6dbf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__base_conversion.txt
@@ -0,0 +1,6 @@
+Quanto é -216 (base 15) em base 6?
+-2103
+-5b8 (em base 15) para base 3
+-1210002
+Quanto é -20112 (base 4) em base 2?
+-1000010110
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder.txt
new file mode 100644
index 0000000000000000000000000000000000000000..7ce08bcf092d45d01f5aec9aa8c98e572cf6d047
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder.txt
@@ -0,0 +1,6 @@
+Calcula o resto da divisão entre 258345 e 708.
+633
+Calcula o resto da divisão entre 5533 e 378.
+241
+Calcula o resto da divisão entre 594877 e 297435.
+7
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f014bbe5521b985276d110691426b2e42abef251
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__div_remainder_composed.txt
@@ -0,0 +1,6 @@
+Supõe 4*a - 94 = 1014. Qual é o resto da divisão entre a e 35?
+32
+Supõe 34 = -5*f + 174. Seja u(q) = -q**3 - 10*q**2 - 6*q + 36. Qual é o resto da divisão entre f e u(-9)?
+1
+Calcula o resto da divisão entre 454 e 638/(-1595) - -257*(-2)/(-10).
+46
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f78eb3db0c0b8512d04f45055238d5c9e7425a13
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd.txt
@@ -0,0 +1,6 @@
+Calcula o maior fator comum  de 418 e 3167186.
+418
+Calcula o maior divisor comum  de 15834 e 343.
+7
+Qual é o maior divisor comum  de 5360 e 191360?
+80
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1f4f61e313097a5f0e78f64f872699a6eb639581
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__gcd_composed.txt
@@ -0,0 +1,6 @@
+Seja p(w) a terceira derivada de w**5/30 + 3*w**4/8 + 4*w**3/3 + 6*w**2. Seja u, p(-6). Qual é o maior divisor comum  de 182 e u?
+26
+Supõe 55*u = 3*a + 51*u - 5174, 0 = 4*a + u - 6886. Qual é o maior divisor comum  de 42 e a?
+42
+Assume que c é (1/(-2))/((-9)/(-1746)). Seja z = 29 + c. Seja b = -40 - z. Qual é o maior fator comum  de b e 14?
+14
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e26ce7bbe28297b1f23b1798d4ad20605d9bd2b2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor.txt
@@ -0,0 +1,6 @@
+Será 7 um fator de 372950?
+False
+Será 1051916 um múltiplo de 446?
+False
+Será que 144 divide 51500?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..72b7ecb9836c3a2d23a103ab8065535ed3acc949
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_factor_composed.txt
@@ -0,0 +1,6 @@
+Supõe -30*m = -37*m + 35. Seja z = -7 + 12. Supõe m*f = f - 12, z*f - 39 = -3*q. Será que 4 divide q?
+False
+Seja c(a) = a**2 - 8*a + 7. Seja k(y) = y**2 - 13*y - 4. Seja t, k(14). Será c(t) um múltiplo de 9?
+True
+Seja v = -677 - -817. Será 28 um fator de v?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e55ba0575a39c92770a5e54af3c4cc538276660c
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime.txt
@@ -0,0 +1,6 @@
+Será 2081537 um número primo?
+True
+Será 82103701 compósito?
+True
+Será 31276157 compósito?
+True
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..4092582d75c7718ce33aba1a5bb376f7d53e253d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__is_prime_composed.txt
@@ -0,0 +1,6 @@
+Supõe 6*x - 34613 = 68305. Será x compósito?
+True
+Supõe 4 = 5*n + 2*a + 34, -2*n = 5*a - 9. Assume que v é 1/(1/n)*-1. Seja i = 13 + v. Será i um número primo?
+False
+Supõe -78*w + 882 = -72*w. Supõe 149*u - 354 = w*u. Será u um número primo?
+False
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1681e5663cba18dedb3dbb3d15076a79000ed0e8
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm.txt
@@ -0,0 +1,6 @@
+Calcula o menor múltiplo comum de 68 e 50218.
+100436
+Calcula o denominador comum de 59/91542 e 45.
+91542
+Qual é o denominador comum de 77/1752351 e 59/27?
+15771159
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..f35fca6a38034b11b51ab92877f30b69ac4e3147
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__lcm_composed.txt
@@ -0,0 +1,6 @@
+Seja g = 259/8 - 23. Qual é o denominador comum de g e (70/4*72/(-3750))/((-4)/10)?
+200
+Assume que w é (-81)/(-15)*614/(-2). Seja u = -1650 - w. Encontra o denominador comum de u e -47/3.
+15
+Seja o(k) = 3*k**2 - k - 9*k - 12*k**2 + k - k**3 - 3. Seja x, o(-8). Calcula o menor múltiplo comum de 4 e x*3/((-9)/(-6)).
+20
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b51307bceec3570618057fc34651105beadc3b43
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors.txt
@@ -0,0 +1,6 @@
+Enumera os fatores primos de 1254566.
+2, 17, 36899
+Enumera os fatores primos de 30762.
+2, 3, 1709
+Enumera os fatores primos de 453538.
+2, 226769
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a8df815d8faed4653785978600ee7405895f9e6f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__list_prime_factors_composed.txt
@@ -0,0 +1,6 @@
+Seja u(r) = 2*r - 4. Seja d, u(4). Supõe 0 = -3*h - 4*l - 1091, h - 2*l = -d*l - 367. Enumera os fatores primos de ((-4)/(-6))/((-14)/h).
+17
+Seja i = 53 - 63. Quais são os fatores primos (-719)/(-5) - (32/i - -3)?
+2, 3
+Seja t = 92 + -65. Supõe 0 = 5*m - 5*p - 1000, -4*m - 2*p = -t - 803. Enumera os fatores primos de m.
+5, 41
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value.txt
new file mode 100644
index 0000000000000000000000000000000000000000..db19514e7d2457fd75b43b2f008285991a22b55d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value.txt
@@ -0,0 +1,6 @@
+Qual é o dígito das milhares de 1873939?
+3
+Qual é o dígito das centenas de 754650?
+6
+Qual é o dígito das unidades de 13088799?
+9
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a7b1b017d673a165203e661e4a77212b1076ac64
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__place_value_composed.txt
@@ -0,0 +1,6 @@
+Seja n = -17 + 27. Seja y(q) = q**3 + 6*q**2 + 4*q - 1. Seja m, y(-5). Supõe m*c = 4*a - 32, -46 - n = -4*a - 4*c. Qual é o dígito das dezenas de a?
+1
+Supõe 4*b = 4 + 12. Supõe 0*i - i - 3 = b*r, -r - 9 = 3*i. Supõe r = -p - 0*p + 27. Qual é o dígito das dezenas de p?
+2
+Seja t(s) = -115*s. Seja d = -56 + 55. Qual é o dígito das dezenas de t(d)?
+1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number.txt
new file mode 100644
index 0000000000000000000000000000000000000000..e23abf0c11ecab21ba1cba765bac5255dae960bd
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number.txt
@@ -0,0 +1,6 @@
+Arredonda 324.582 para as dez mais próxima.
+320
+Quanto é 0.0008865303 arredondado para as casa decimal das 6s?
+0.000887
+Quanto é -55971.03 arredondado para as um milhar mais próxima?
+-56000
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..1a9bf6115ef054e5ffe61f17a69e56660435c7de
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/numbers__round_number_composed.txt
@@ -0,0 +1,6 @@
+Seja w = -1 + -7. Seja b = 17 + w. Seja j = 9.00000031 - b. Arredonda j para as casa decimal das 7s.
+0.0000003
+Seja l = -35 - -35.019. Seja a = 0.0190112 - l. Arredonda a para as casa decimal das seiss.
+0.000011
+Seja g = -97.0007 + 97. Quanto é g arredondado para as casa decimal das trêss?
+-0.001
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__add.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__add.txt
new file mode 100644
index 0000000000000000000000000000000000000000..c3be14519a2637551378b42753c52e3da3c341fb
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__add.txt
@@ -0,0 +1,6 @@
+Seja a(h) = -h**3 + h**2. Seja k = -20 + 24. Supõe 5 = -q - k*q. Seja u(v) = 7*v**3 - 3*v**2 + 1. Determina q*u(m) - 5*a(m).
+-2*m**3 - 2*m**2 - 1
+Seja z(n) = 7*n + 2. Supõe -13*v + 400 = 465. Seja o(k) a terceira derivada de 0*k + 0 - 1/2*k**3 - 1/3*k**4 - k**2. Obtém v*z(j) - 4*o(j).
+2 - 3*j
+Seja o(b) = 3*b - 2. Supõe 0*a + 4*a - 42 = -3*d, -2*d = -3*a - 45. Seja c(t) = -d*t + 36*t - 19*t. O que é 2*c(x) + o(x)?
+x - 2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__coefficient_named.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__coefficient_named.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b2567164de8e33a1e383bb27ed76b07a9a2814f3
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__coefficient_named.txt
@@ -0,0 +1,6 @@
+Expressa 2*t - 2*t + t**4 + (-t + t**2 + t)*(-120 + 120 - 43*t)*(-t - t - 3*t) na forma p + b*t**2 + i*t**4 + w*t**3 + c*t e dá i.
+216
+Expressa -116*f**2 + 15*f**2 + 13 - 13 como m + i*f**2 + r*f e dá i.
+-101
+Expressa 4*v**4 + 0*v - 9*v**4 + 4*v + 3*v**3 + 2 - 2*v**3 + 7*v**4 na forma o*v**3 + a*v + j*v**2 + c + z*v**4 e dá c.
+2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__collect.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__collect.txt
new file mode 100644
index 0000000000000000000000000000000000000000..8b8d4d3e2acf9c57aaede7b6fbb428df1851890e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__collect.txt
@@ -0,0 +1,6 @@
+Coleta os termos em 1000*z**3 + 480*z**3 + z + z - 2*z.
+1480*z**3
+Coleta os termos em -42*j**2 + 452 - 27*j**2 - 452.
+-69*j**2
+Coleta os termos em 58*m**3 + 115 + 19 + 81 - 68*m**3.
+215 - 10*m**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__compose.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__compose.txt
new file mode 100644
index 0000000000000000000000000000000000000000..edf3be5052dfa0d99376b2fb2e2529e7db99863a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__compose.txt
@@ -0,0 +1,6 @@
+Seja w(x) = -108*x - 239*x + 1152*x - 328*x. Seja f(b) = 2*b. O que é f(w(j))?
+954*j
+Seja h(j) = 17 - 186*j. Seja p(u) = 23*u - 2. Seja b(w) = 6*h(w) + 51*p(w). Seja z(n) = -3*n. Calcula b(z(f)).
+-171*f
+Seja i(a) = -a**2 + 4*a. Seja d(z) = 3*z. Seja b(f) = -4*d(f) + 3*i(f). Seja q(o) = -5*o. O que é b(q(h))?
+-75*h**2
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate.txt
new file mode 100644
index 0000000000000000000000000000000000000000..15767e0be2ac99728fcd9d23fff281bcf47195ad
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate.txt
@@ -0,0 +1,6 @@
+Seja s(l) = 2*l**2 + 286*l + 3832. Calcula s(-15).
+-8
+Seja t(v) = -4*v**2 + 325*v + 244. Determina t(82).
+-2
+Seja c(u) = -u**2 + 18*u + 402. Determina c(-13).
+-1
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate_composed.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate_composed.txt
new file mode 100644
index 0000000000000000000000000000000000000000..55320fdefe6ae0dbc94f74a785fe9055f328b714
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__evaluate_composed.txt
@@ -0,0 +1,6 @@
+Seja m(z) = 1 - 2*z. Supõe -x = 5*f - 2*f - 1, 2*x - 4 = -4*f. Supõe 0 = -4*w + 6*w - x. O que é m(w)?
+-3
+Seja p(b) = -b**3 + 10*b**2 - 9*b + 3. Seja i, p(9). Seja y(r) a terceira derivada de 0*r + 2*r**2 - 1/12*r**4 - 1/3*r**i + 0. Determina y(3).
+-8
+Seja r(t) = t. Seja p = 529 - 525. Obtém r(p).
+4
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__expand.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__expand.txt
new file mode 100644
index 0000000000000000000000000000000000000000..a2ab5fc2d50743e8c62d5f32687692fe43d22c42
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__expand.txt
@@ -0,0 +1,6 @@
+Expande (-26 + 5 + 1)*(-4 + 4 + 3*b).
+-60*b
+Expande ((-3*a + a - a)*(-5*a**2 + 623 - 623) + 0 - 2*a**3 + 0)*(5*a + 0*a - 2*a).
+39*a**4
+Expande (n**2 - 3*n**2 + 3*n**2)*(-8 - 23*n - 8 + 16).
+-23*n**3
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__simplify_power.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__simplify_power.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ec329c200ba2a5d1c39958f300860c07543823d8
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/polynomials__simplify_power.txt
@@ -0,0 +1,6 @@
+Simplifica ((((y*y/(y/y**(-8)*y)*y)/y)/y)/y**3)/((y**8*y)/(y/(y/(y**11*y)*y))) assumindo que y é positiva.
+y**(-10)
+Simplifica (((v**(-11)/v)/(((v*v*(v/v**12)/v)/v)/v))**27)**(-1/42) assumindo que v é positiva.
+1
+Simplifica (q/(q/(q/(q*(((q/(q*q**6*q))/q)/q)/q))))**12*(q**0/q)**(-1/5) assumindo que q é positiva.
+q**(601/5)
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_level_set.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_level_set.txt
new file mode 100644
index 0000000000000000000000000000000000000000..6844d3f6cb76a61a22d9323494323959395e27cf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_level_set.txt
@@ -0,0 +1,6 @@
+Três caractéres são escolhidos sem reposição de syaasyattyttaayyt. Qual é a probabilidade de(a) escolher 2 y e 1 s?
+1/34
+Qual é a probabilidade de(a) escolher 1 v e 1 r quando dois caractéres são escolhidos sem reposição de lvllrlvrlrv?
+9/55
+Três caractéres são escolhidos sem reposição de vvvqqrrvvrqvrvrq. Dá a probabilidade de(a) escolher 1 q e 2 v.
+3/20
diff --git a/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_sequence.txt b/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_sequence.txt
new file mode 100644
index 0000000000000000000000000000000000000000..b658c96b748c4549fe372c9d84110b318e4c3fb8
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/ds/train-medium/probability__swr_p_sequence.txt
@@ -0,0 +1,6 @@
+Quatro caractéres são escolhidos sem reposição de {y: 4, r: 9, m: 1}. Dá a probabilidade de(a) sequência myrr.
+12/1001
+Três caractéres são escolhidos sem reposição de {n: 5, v: 3, t: 1}. Qual é a probabilidade de(a) sequência tnn?
+5/126
+Qual é a probabilidade de(a) sequência lls quando três caractéres são escolhidos sem reposição de {l: 2, s: 1}?
+1/3
diff --git a/mathematics_dataset/mathematics_dataset/example.py b/mathematics_dataset/mathematics_dataset/example.py
new file mode 100644
index 0000000000000000000000000000000000000000..1aa3b27fc428743248234a18c3637a1e3242f43a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/example.py
@@ -0,0 +1,59 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Containers for "[example] problems" (i.e., question/answer) pairs."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+
+from util import composition
+
+
+def question(context, template, **kwargs):
+  """Makes a question, using the given context and template.
+
+  The format is similar to that for python's `format` function, for example:
+
+  ```
+  question(context, 'What is {} plus {p} over {q}?', 2, p=3, q=4)
+  ```
+
+  The main difference between this and the standard python formatting is that
+  this understands `Entity`s in the arguments, and will do appropriate expansion
+  of text and prefixing of their descriptions.
+
+  Arguments:
+    context: Instance of `composition.Context`, for extracting entities needed
+        for describing the problem.
+    template: A string, like "Calculate the value of {exp}.".
+    **kwargs: A dictionary mapping arguments to values, e.g.,
+        `{'exp': sympy.Add(2, 3, evaluate=False)}`.
+
+  Returns:
+    String.
+  """
+  assert isinstance(context, composition.Context)
+  assert isinstance(template, str)
+  prefix, kwargs = composition.expand_entities(context, **kwargs)
+  if prefix:
+    prefix += ' '
+  return prefix + template.format(**kwargs)
+
+
+Problem = collections.namedtuple('Problem', ('question', 'answer'))
+
+
diff --git a/mathematics_dataset/mathematics_dataset/generate.py b/mathematics_dataset/mathematics_dataset/generate.py
new file mode 100644
index 0000000000000000000000000000000000000000..22565e10f5f7cfb7dfab1737b7f73846135874ec
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/generate.py
@@ -0,0 +1,172 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Prints to stdout different curriculum questions."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+import textwrap
+
+# Dependency imports
+from absl import app
+from absl import flags
+from absl import logging
+import generate_settings
+from modules import modules
+import six
+from six.moves import range
+
+
+FLAGS = flags.FLAGS
+
+flags.DEFINE_string('filter', '', 'restrict to matching module names')
+flags.DEFINE_integer('per_train_module', 10, 'Num of examples per train module')
+flags.DEFINE_integer('per_test_module', 10, 'Num of examples per test module')
+flags.DEFINE_bool('show_dropped', False, 'Whether to print dropped questions')
+
+
+filtered_modules = collections.OrderedDict([])
+counts = {}
+
+
+def _make_entropy_fn(level, num_levels):
+  """This returns a function that returns a subrange of entropy.
+
+  E.g., if level=1 (medium) and num_levels=3, then the returned function will
+  map the range [x, x + y] to [x + y/3, x + 2y/3].
+
+  Args:
+    level: Integer in range [0, num_levels - 1].
+    num_levels: Number of difficulty levels.
+
+  Returns:
+    Function to restrict entropy range.
+  """
+  lower = level / num_levels
+  upper = (level + 1) / num_levels
+  def modify_entropy(range_):
+    assert len(range_) == 2
+    length = range_[1] - range_[0]
+    return (range_[0] + lower * length, range_[0] + upper * length)
+  return modify_entropy
+
+
+def _filter_and_flatten(modules_):
+  """Returns flattened dict, filtered according to FLAGS."""
+  flat = collections.OrderedDict()
+
+  def add(submodules, prefix=None):
+    for key, module_or_function in six.iteritems(submodules):
+      full_name = prefix + '__' + key if prefix is not None else key
+      if isinstance(module_or_function, dict):
+        add(module_or_function, full_name)
+      else:
+        if FLAGS.filter not in full_name:
+          continue
+        flat[full_name] = module_or_function
+
+  add(modules_)
+
+  # Make sure list of modules are in deterministic order. This is important when
+  # generating across multiple machines.
+  flat = collections.OrderedDict(
+      [(key, flat[key]) for key in sorted(six.iterkeys(flat))])
+
+  return flat
+
+
+def init_modules(train_split=False):
+  """Inits the dicts containing functions for generating modules."""
+  if filtered_modules:
+    return  # already initialized
+
+  all_modules = collections.OrderedDict([])
+  if train_split:
+    all_modules['train-easy'] = modules.train(_make_entropy_fn(0, 3))
+    all_modules['train-medium'] = modules.train(_make_entropy_fn(1, 3))
+    all_modules['train-hard'] = modules.train(_make_entropy_fn(2, 3))
+  else:
+    all_modules['train'] = modules.train(_make_entropy_fn(0, 1))
+
+  all_modules['interpolate'] = modules.test()
+  all_modules['extrapolate'] = modules.test_extra()
+
+  counts['train'] = FLAGS.per_train_module
+  counts['train-easy'] = FLAGS.per_train_module // 3
+  counts['train-medium'] = FLAGS.per_train_module // 3
+  counts['train-hard'] = FLAGS.per_train_module // 3
+  counts['interpolate'] = FLAGS.per_test_module
+  counts['extrapolate'] = FLAGS.per_test_module
+
+  for regime_, modules_ in six.iteritems(all_modules):
+    filtered_modules[regime_] = _filter_and_flatten(modules_)
+
+
+def sample_from_module(module):
+  """Samples a problem, ignoring samples with overly long questions / answers.
+
+  Args:
+    module: Callable returning a `Problem`.
+
+  Returns:
+    Pair `(problem, num_dropped)`, where `problem` is an instance of `Problem`
+    and `num_dropped` is an integer >= 0 indicating the number of samples that
+    were dropped.
+  """
+  num_dropped = 0
+  while True:
+    problem = module()
+    question = str(problem.question)
+    if len(question) > generate_settings.MAX_QUESTION_LENGTH:
+      num_dropped += 1
+      if FLAGS.show_dropped:
+        logging.warning('Dropping question: %s', question)
+      continue
+    answer = str(problem.answer)
+    if len(answer) > generate_settings.MAX_ANSWER_LENGTH:
+      num_dropped += 1
+      if FLAGS.show_dropped:
+        logging.warning('Dropping question with answer: %s', answer)
+      continue
+    return problem, num_dropped
+
+
+def main(unused_argv):
+  """Prints Q&As from modules according to FLAGS.filter."""
+  init_modules()
+
+  text_wrapper = textwrap.TextWrapper(
+      width=80, initial_indent=' ', subsequent_indent='  ')
+
+  for regime, flat_modules in six.iteritems(filtered_modules):
+    per_module = counts[regime]
+    for module_name, module in six.iteritems(flat_modules):
+      # These magic print constants make the header bold.
+      print('\033[1m{}/{}\033[0m'.format(regime, module_name))
+      num_dropped = 0
+      for _ in range(per_module):
+        problem, extra_dropped = sample_from_module(module)
+        num_dropped += extra_dropped
+        text = text_wrapper.fill(
+            '{}  \033[92m{}\033[0m'.format(problem.question, problem.answer))
+        print(text)
+      if num_dropped > 0:
+        logging.warning('Dropped %d examples', num_dropped)
+
+
+if __name__ == '__main__':
+  app.run(main)
diff --git a/mathematics_dataset/mathematics_dataset/generate_settings.py b/mathematics_dataset/mathematics_dataset/generate_settings.py
new file mode 100644
index 0000000000000000000000000000000000000000..29e0c69bf58dc8c15ac5e70b71afe031a7311b2f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/generate_settings.py
@@ -0,0 +1,32 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Settings for generation."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import string
+
+MAX_QUESTION_LENGTH = 160
+MAX_ANSWER_LENGTH = 30
+QUESTION_CHARS = (
+    ['', ' '] + list(string.ascii_letters + string.digits + string.punctuation))
+EMPTY_INDEX = QUESTION_CHARS.index('')
+NUM_INDICES = len(QUESTION_CHARS)
+CHAR_TO_INDEX = {char: index for index, char in enumerate(QUESTION_CHARS)}
+INDEX_TO_CHAR = {index: char for index, char in enumerate(QUESTION_CHARS)}
+
diff --git a/mathematics_dataset/mathematics_dataset/generate_test.py b/mathematics_dataset/mathematics_dataset/generate_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..ae28822b3236dd7e9e4885b8a93b87bc637aca4b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/generate_test.py
@@ -0,0 +1,47 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.generate."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from absl.testing import parameterized
+import generate
+import six
+from six.moves import range
+
+
+class GenerateTest(parameterized.TestCase):
+
+  def testMakeEntropyFn(self):
+    entropy_full = generate._make_entropy_fn(0, 1)
+    self.assertEqual(entropy_full((2, 3)), (2, 3))
+    entropy_third = generate._make_entropy_fn(2, 3)
+    self.assertEqual(entropy_third((3, 6)), (5, 6))
+
+  @parameterized.parameters('train', 'interpolate', 'extrapolate')
+  def testGenerate(self, regime):
+    generate.init_modules()
+    for module in six.itervalues(generate.filtered_modules[regime]):
+      for _ in range(3):
+        question = module()
+        str(question)
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/generate_to_file.py b/mathematics_dataset/mathematics_dataset/generate_to_file.py
new file mode 100644
index 0000000000000000000000000000000000000000..6e73154299c005c0fafd0307876c3441a89e5183
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/generate_to_file.py
@@ -0,0 +1,79 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Example of how to write generated questions to text files.
+
+Given an output directory, this will create the following subdirectories:
+
+*   train-easy
+*   train-medium
+*   train-hard
+*   interpolate
+*   extrapolate
+
+and populate each of these directories with a text file for each of the module,
+where the text file contains lines alternating between the question and the
+answer.
+
+Passing --train_split=False will create a single output directory 'train' for
+training data.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+
+# Dependency imports
+from absl import app
+from absl import flags
+from absl import logging
+import generate
+import six
+from six.moves import range
+
+FLAGS = flags.FLAGS
+
+flags.DEFINE_string('output_dir', None, 'Where to write output text')
+flags.DEFINE_boolean('train_split', True,
+                     'Whether to split training data by difficulty')
+flags.mark_flag_as_required('output_dir')
+
+
+def main(unused_argv):
+  generate.init_modules(FLAGS.train_split)
+
+  output_dir = os.path.expanduser(FLAGS.output_dir)
+  if os.path.exists(output_dir):
+    logging.fatal('output dir %s already exists', output_dir)
+  logging.info('Writing to %s', output_dir)
+  os.makedirs(output_dir)
+
+  for regime, flat_modules in six.iteritems(generate.filtered_modules):
+    regime_dir = os.path.join(output_dir, regime)
+    os.mkdir(regime_dir)
+    per_module = generate.counts[regime]
+    for module_name, module in six.iteritems(flat_modules):
+      path = os.path.join(regime_dir, module_name + '.txt')
+      with open(path, 'w',encoding='utf-8') as text_file:
+        for _ in range(per_module):
+          problem, _ = generate.sample_from_module(module)
+          text_file.write(str(problem.question) + '\n')
+          text_file.write(str(problem.answer) + '\n')
+      logging.info('Written %s', path)
+
+
+if __name__ == '__main__':
+  app.run(main)
diff --git a/mathematics_dataset/mathematics_dataset/modules/__init__.py b/mathematics_dataset/mathematics_dataset/modules/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..d780829a1f5101333577a1ca9adcb960529213f1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/__init__.py
@@ -0,0 +1,13 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/mathematics_dataset/mathematics_dataset/modules/algebra.py b/mathematics_dataset/mathematics_dataset/modules/algebra.py
new file mode 100644
index 0000000000000000000000000000000000000000..f1f15932dbb95fba3275b2df433d3c947ad238c7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/algebra.py
@@ -0,0 +1,378 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Algebra-related questions, e.g., "Solve 1 + x = 2."."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import random
+
+# Dependency imports
+import example
+from sample import linear_system
+from sample import number
+from sample import ops
+from sample import polynomials
+from util import composition
+from util import display
+import numpy as np
+from six.moves import range
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+_ENTROPY_EXTRAPOLATE = (12, 12)
+
+# In generating a polynomial with real roots (where the roots are generated
+# sequentially), this is the probability of taking a previous root, thus giving
+# at least one repeated root, rather than sampling a new number. The value is
+# somewhat arbitrary, but gives a "medium probability" of seeing a repeated root
+# for lowish degree polynomials.
+_POLY_PROBABILITY_REPEATED_ROOT = 0.2
+
+
+def _make_modules(entropy):
+  """Returns modules given "difficulty" parameters."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  sample_args_composed = composition.PreSampleArgs(2, 4, *entropy)
+
+  return {
+      # Solving equations:
+      'polynomial_roots': functools.partial(
+          polynomial_roots, None, sample_args_pure),
+      'polynomial_roots_composed': functools.partial(
+          polynomial_roots, None, sample_args_composed),
+      'linear_1d': functools.partial(
+          solve_linear_1d, None, sample_args_pure),
+      'linear_1d_composed': functools.partial(
+          solve_linear_1d, None, sample_args_composed),
+      'linear_2d': functools.partial(
+          solve_linear_2d, None, sample_args_pure),
+      'linear_2d_composed': functools.partial(
+          solve_linear_2d, None, sample_args_composed),
+
+      # Sequences:
+      'sequence_next_term': functools.partial(sequence_next_term, *entropy),
+      'sequence_nth_term': functools.partial(sequence_nth_term, *entropy),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(entropy_fn(_ENTROPY_TRAIN))
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(_ENTROPY_INTERPOLATE)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *_ENTROPY_EXTRAPOLATE)
+  return {
+      'polynomial_roots_big': functools.partial(
+          polynomial_roots, None, sample_args_pure),
+  }
+
+
+def _sample_roots(entropy):
+  """Generates `num_distinct + num_repeated` polynomial roots."""
+  num_roots = random.randint(2, 5)
+
+  num_repeated = np.random.binomial(
+      num_roots - 1, _POLY_PROBABILITY_REPEATED_ROOT)
+  # Slight hack: don't allow all the roots to be repeated when the entropy is
+  # high, as this can create very large coefficients.
+  if entropy > 4:
+    num_repeated = min(num_repeated, int(num_roots / 2))
+
+  num_distinct = num_roots - num_repeated
+
+  entropies = entropy * np.random.dirichlet(np.ones(num_distinct))
+
+  roots = []
+
+  for root_entropy in entropies:
+    # Generates a root with small probability of being rational.
+    # (Otherwise when we multiply out the denominators, we get really large
+    # coefficients in our polynomial.)
+    if random.random() < 0.1:
+      root = number.non_integer_rational(root_entropy, True)
+    else:
+      root = number.integer(root_entropy, True)
+    roots.append(root)
+
+  for _ in range(num_repeated):
+    roots.append(random.choice(roots[:num_distinct]))
+
+  return roots
+
+
+def _polynomial_coeffs_with_roots(roots, scale_entropy):
+  """Returns a polynomial with the given roots.
+
+  The polynomial is generated by expanding product_{root in roots} (x - root),
+  and then (1) scaling by the coefficients so they are all integers with lcm 1,
+  and then (2) further scaling the coefficients by a random integer or rational
+  with `scale_entropy` digits.
+
+  Args:
+    roots: List of values.
+    scale_entropy: Float; entropy of the random coefficient scaling.
+
+  Returns:
+    List of coefficients `coeffs`, such that `coeffs[i]` is the coefficient of
+    variable ** i.
+  """
+  variable = sympy.Symbol('x')  # doesn't matter, only use coefficients
+  polynomial = sympy.Poly(sympy.prod([variable - root for root in roots]))
+  coeffs_reversed = polynomial.all_coeffs()
+  assert len(coeffs_reversed) == len(roots) + 1
+  coeffs = list(reversed(coeffs_reversed))
+  # Multiply terms to change rationals to integers, and then maybe reintroduce.
+  lcm = sympy.lcm([sympy.denom(coeff) for coeff in coeffs])
+  if scale_entropy > 0:
+    while True:
+      scale = number.integer_or_rational(scale_entropy, signed=True)
+      if scale != 0:
+        break
+  else:
+    scale = 1
+  return [coeff * scale * lcm for coeff in coeffs]
+
+
+def polynomial_roots(value, sample_args, context=None):
+  """E.g., "Solve 2*x**2 - 18 = 0."."""
+  del value  # not currently used
+  # is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  scale_entropy = min(entropy / 2, 1)
+
+  roots = _sample_roots(entropy - scale_entropy)
+  solutions = sorted(list(sympy.FiniteSet(*roots)))
+  coeffs = _polynomial_coeffs_with_roots(roots, scale_entropy)
+  (polynomial_entity,) = context.sample(
+      sample_args, [composition.Polynomial(coeffs)])
+
+  if random.choice([False, True]):
+    # Ask for explicit roots.
+    if len(solutions) == 1:
+      answer = solutions[0]
+    else:
+      answer = display.NumberList(solutions)
+
+    if polynomial_entity.has_expression():
+      equality = ops.Eq(polynomial_entity.expression, 0)
+      variable = polynomial_entity.polynomial_variables[0]
+    else:
+      variable = sympy.Symbol(context.pop())
+      equality = ops.Eq(polynomial_entity.handle.apply(variable), 0)
+    template = random.choice([
+      'Seja {equality}. Qual é {variable}?',
+      'Seja {equality}. Calcula {variable}.',
+      'Supõe {equality}. Qual é {variable}?',
+      'Supõe {equality}. Calcula {variable}?',
+      'Qual é o valor de {variable} em {equality}?',
+      'Resolve {equality} em relação a {variable}.',
+      'Encontra {variable} tal que {equality}.',
+      'Encontra {variable}, dado que {equality}.',
+      'Determina {variable} para que {equality}.',
+      'Determina {variable}, dado que {equality}.',
+      'Resolve {equality}.'
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, equality=equality, variable=variable),
+        answer=answer)
+  else:
+    if polynomial_entity.has_expression():
+      expression = polynomial_entity.expression
+      variable = polynomial_entity.polynomial_variables[0]
+    else:
+      variable = sympy.Symbol(context.pop())
+      expression = polynomial_entity.handle.apply(variable)
+    factored = sympy.factor(
+        polynomials.coefficients_to_polynomial(coeffs, variable))
+    template = random.choice([
+        'Factoriza {expression}.',
+    ])
+    return example.Problem(
+        question=example.question(context, template, expression=expression),
+        answer=factored)
+
+
+def _solve_linear_system(degree, value, sample_args, context=None):
+  """Solve linear equations."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  solutions = []
+  if value is not None:
+    solutions.append(value)
+
+  extra_solutions_needed = degree - len(solutions)
+  if extra_solutions_needed > 0:
+    entropies = (entropy / 4) * np.random.dirichlet(
+        np.ones(extra_solutions_needed))
+    entropies = np.maximum(1, entropies)  # min per-solution entropy
+    entropy -= sum(entropies)
+    solutions += [number.integer(solution_entropy, True)
+                  for solution_entropy in entropies]
+  entropy = max(1, entropy)
+
+  variables = [sympy.Symbol(context.pop()) for _ in range(degree)]
+
+  solution_index = 0
+  # If we're going to be creating a linear system with constants to replace by
+  # handles from other modules, then we need a linear system with constants
+  # occurring. Very occasionally this can fail to happen, e.g., "x = -x";
+  # normally this while loop will only see one iteration.
+  while True:
+    equations = linear_system.linear_system(
+        variables=variables, solutions=solutions, entropy=entropy,
+        non_trivial_in=solution_index)
+    constants = ops.number_constants(equations)
+    if sample_args.num_modules <= 1 or constants:
+      break
+
+  context.sample_by_replacing_constants(sample_args, equations)
+
+  variable = variables[solution_index]
+  answer = solutions[solution_index]
+
+  equations = ', '.join([str(equation) for equation in equations])
+
+  if is_question:
+    template = random.choice([
+        'Resolve {equations} para {variable}.',
+    ])
+    return example.Problem(
+        example.question(
+            context, template, equations=equations,
+            variable=variable),
+        answer)
+  else:
+    return composition.Entity(
+        context=context,
+        value=answer,
+        description='Supõe {equations}.',
+        handle=variable,
+        equations=equations)
+
+
+@composition.module(number.is_integer)
+def solve_linear_1d(*args, **kwargs):
+  return _solve_linear_system(1, *args, **kwargs)
+
+
+@composition.module(number.is_integer)
+def solve_linear_2d(*args, **kwargs):
+  return _solve_linear_system(2, *args, **kwargs)
+
+
+class _PolynomialSequence(object):
+  """A sequence given by a polynomial."""
+
+  def __init__(self, variable, entropy, min_degree=1, max_degree=3):
+    """Initializes a random polynomial sequence.
+
+    Args:
+      variable: Variable to use.
+      entropy: Entropy for polynomial coefficients.
+      min_degree: Minimum order of polynomial.
+      max_degree: Maximum order of polynomial.
+    """
+    self._degree = random.randint(min_degree, max_degree)
+    self._variable = variable
+    polynomial = polynomials.sample_with_small_evaluation(
+        variable=self._variable, degree=self._degree,
+        max_abs_input=self._degree + 2, entropy=entropy)
+    self._sympy = polynomial.sympy()
+
+  @property
+  def min_num_terms(self):
+    """Returns the minimum number of terms to identify the sequence.
+
+    This assumes a human-like prior over types of sequences.
+
+    Returns:
+      Integer >= 1.
+    """
+    return self._degree + 2
+
+  @property
+  def sympy(self):
+    return self._sympy
+
+  def term(self, n):
+    """Returns the `n`th term of the sequence."""
+    return self._sympy.subs(self._variable, n)
+
+
+def sequence_next_term(min_entropy, max_entropy):
+  """E.g., "What is the next term in the sequence 1, 2, 3?"."""
+  entropy = random.uniform(min_entropy, max_entropy)
+  context = composition.Context()
+  variable = sympy.Symbol(context.pop())
+
+  sequence = _PolynomialSequence(variable, entropy)
+  min_num_terms = sequence.min_num_terms
+  num_terms = random.randint(min_num_terms, min_num_terms + 3)
+  sequence_sample = [sequence.term(n + 1) for n in range(num_terms)]
+  sequence_sample = display.NumberList(sequence_sample)
+
+  template = random.choice([
+      'O que se segue em {sequence}?',
+      'O que vem a seguir: {sequence}?',
+      'Qual é o termo seguinte em {sequence}?',
+  ])
+  answer = sequence.term(num_terms + 1)
+
+  return example.Problem(
+      question=example.question(context, template, sequence=sequence_sample),
+      answer=answer)
+
+
+def sequence_nth_term(min_entropy, max_entropy):
+  """E.g., "What is the nth term in the sequence 1, 2, 3?"."""
+  entropy = random.uniform(min_entropy, max_entropy)
+  context = composition.Context()
+  variable = sympy.Symbol(context.pop())
+
+  sequence = _PolynomialSequence(variable, entropy)
+  min_num_terms = sequence.min_num_terms
+  num_terms = random.randint(min_num_terms, min_num_terms + 3)
+  sequence_sample = [sequence.term(n + 1) for n in range(num_terms)]
+  sequence_sample = display.NumberList(sequence_sample)
+
+  template = random.choice([
+      'Qual é termo nº {variable} na sequência {sequence}?',
+  ])
+  answer = sequence.sympy
+
+  return example.Problem(
+      question=example.question(
+          context, template, variable=variable, sequence=sequence_sample),
+      answer=answer)
diff --git a/mathematics_dataset/mathematics_dataset/modules/algebra_test.py b/mathematics_dataset/mathematics_dataset/modules/algebra_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..be4c47a6fca93b3489a81126702f000d53fc58ef
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/algebra_test.py
@@ -0,0 +1,48 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.modules.algebra."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from absl.testing import absltest
+from modules import algebra
+from sample import polynomials
+from six.moves import range
+import sympy
+
+
+class AlgebraTest(absltest.TestCase):
+
+  def testPolynomialCoeffsWithRoots(self):
+    coeffs = algebra._polynomial_coeffs_with_roots([1, 2], scale_entropy=0.0)
+    self.assertEqual(coeffs, [2, -3, 1])
+
+  def testPolynomialRoots(self):
+    variable = sympy.Symbol('x')
+    for _ in range(10):
+      roots = random.sample(list(range(-9, 10)), 3)
+      coeffs = algebra._polynomial_coeffs_with_roots(roots, scale_entropy=10.0)
+      polynomial = polynomials.coefficients_to_polynomial(coeffs, variable)
+      calc_roots = sympy.polys.polytools.real_roots(polynomial)
+      self.assertEqual(calc_roots, sorted(roots))
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/modules/arithmetic.py b/mathematics_dataset/mathematics_dataset/modules/arithmetic.py
new file mode 100644
index 0000000000000000000000000000000000000000..559be777025b3bed887954314763ffc902916380
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/arithmetic.py
@@ -0,0 +1,597 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Arithmetic, e.g., "calculate 2+3"."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import math
+import random
+
+# Dependency imports
+import example
+from sample import arithmetic
+from sample import number
+from sample import ops
+from util import composition
+from util import display
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+_ENTROPY_EXTRAPOLATE = (10, 12)
+
+_ADD_SUB_ENTROPY_TRAIN = (4, 16)
+_ADD_SUB_ENTROPY_INTERPOLATE = (12, 12)
+_ADD_SUB_ENTROPY_EXTRAPOLATE = (16, 20)
+
+# In arithmetic expressions:
+_EXTRAPOLATE_EXTRA_LENGTH = 3
+
+_INT = 'int'
+_INT_OR_RATIONAL = 'rational'
+
+
+def _make_modules(entropy, add_sub_entropy):
+  """Returns modules given "difficulty" parameters."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  add_sub_sample_args_pure = composition.PreSampleArgs(1, 1, *add_sub_entropy)
+
+  # TODO(b/124039105): consider composed modules?
+  return {
+      # Addition and subtraction of integers (and decimals)
+      'add_or_sub': functools.partial(
+          add_or_sub, None, add_sub_sample_args_pure),
+      'add_sub_multiple': functools.partial(
+          add_sub_multiple, _INT, sample_args_pure),
+      'add_or_sub_in_base': functools.partial(
+          add_or_sub_in_base, sample_args_pure),
+
+      # Multiplication and division
+      'mul': functools.partial(mul, None, sample_args_pure),
+      'div': functools.partial(div, None, sample_args_pure),
+      'mul_div_multiple': functools.partial(
+          mul_div_multiple, _INT_OR_RATIONAL, sample_args_pure),
+
+      # All together!
+      'mixed': functools.partial(mixed, _INT_OR_RATIONAL, sample_args_pure),
+
+      # And some other arithmetic-related stuff.
+      'nearest_integer_root': functools.partial(
+          nearest_integer_root, sample_args_pure),
+      'simplify_surd': functools.partial(simplify_surd, None, sample_args_pure),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(
+      entropy=entropy_fn(_ENTROPY_TRAIN),
+      add_sub_entropy=entropy_fn(_ADD_SUB_ENTROPY_TRAIN))
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(
+      entropy=_ENTROPY_INTERPOLATE,
+      add_sub_entropy=_ADD_SUB_ENTROPY_INTERPOLATE)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *_ENTROPY_EXTRAPOLATE)
+  add_sub_sample_args_pure = composition.PreSampleArgs(
+      1, 1, *_ADD_SUB_ENTROPY_EXTRAPOLATE)
+
+  train_length = arithmetic.length_range_for_entropy(_ENTROPY_TRAIN[1])[1]
+  def extrapolate_length():
+    return random.randint(
+        train_length + 1, train_length + _EXTRAPOLATE_EXTRA_LENGTH)
+
+  def add_sub_multiple_longer():
+    return add_sub_multiple(_INT, sample_args_pure, length=extrapolate_length())
+  def mul_div_multiple_longer():
+    return mul_div_multiple(_INT, sample_args_pure, length=extrapolate_length())
+  def mixed_longer():
+    return mixed(_INT, sample_args_pure, length=extrapolate_length())
+
+  return {
+      'add_or_sub_big': functools.partial(
+          add_or_sub, None, add_sub_sample_args_pure),
+      'mul_big': functools.partial(mul, None, sample_args_pure),
+      'div_big': functools.partial(div, None, sample_args_pure),
+      'add_sub_multiple_longer': add_sub_multiple_longer,
+      'mul_div_multiple_longer': mul_div_multiple_longer,
+      'mixed_longer': mixed_longer,
+  }
+
+
+def _value_sampler(value):
+  """Returns sampler (e.g., number.integer) appropriate for `value`."""
+  if value == _INT or number.is_integer(value):
+    return functools.partial(number.integer, signed=True)
+  if value == _INT_OR_RATIONAL or isinstance(value, sympy.Rational):
+    return functools.partial(number.integer_or_rational, signed=True)
+  if isinstance(value, display.Decimal):
+    return functools.partial(number.integer_or_decimal, signed=True)
+  raise ValueError('Unrecognized value {} of type {}'
+                   .format(value, type(value)))
+
+
+def _add_question_or_entity(context, p, q, is_question):
+  """Generates entity or question for adding p + q."""
+  value = p.value + q.value
+
+  if is_question:
+    template = random.choice([
+        '{p} + {q}',
+        '{p}+{q}',
+        'Determina {p} + {q}.',
+        'Adiciona {p} a {q}.',
+        'Junta {p} e {q}.',
+        'Soma {p} e {q}.',
+        'O total de {p} mais {q}.',
+        'Acresce {p} a {q}.',
+        'Quanto é {p} mais {q}?',
+        'Calcula {p} + {q}.',
+        'Quanto é {p} + {q}?',
+    ])
+    return example.Problem(
+        question=example.question(context, template, p=p, q=q),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        description='Seja {self} = {p} + {q}.',
+        p=p, q=q)
+
+
+def _sub_question_or_entity(context, p, q, is_question):
+  """Generates entity or question for subtraction p - q."""
+  value = p.value - q.value
+
+  if is_question:
+    templates = [
+        '{p} - {q}',
+        'Determina {p} - {q}.',
+        'Quanto é {p} menos {q}?',
+        'Quanto fica {p} retirando {q}?',
+        'Quanto é {q} retirado de {p}?',
+        'Subtrai {q} de {p}.',
+        'Calcula {p} - {q}.',
+        'Quanto é {p} - {q}?',
+    ]
+    if sympy.Ge(p.value, q.value):
+      # We calculate p - q, so the difference (|p - q|) is the correct answer.
+      for adjective in ['distância', 'diferença']:
+        for pair in ['{p} e {q}', '{q} e {p}']:
+          templates.append('Qual é a {} entre {}?'.format(adjective, pair))
+    template = random.choice(templates)
+    return example.Problem(
+        question=example.question(context, template, p=p, q=q),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        description='Seja {self} = {p} - {q}.',
+        p=p, q=q)
+
+
+def _entropy_for_pair(entropy):
+  entropy_1 = max(1, random.uniform(0, entropy))
+  entropy_2 = max(1, entropy - entropy_1)
+  return entropy_1, entropy_2
+
+
+@composition.module(number.is_integer_or_rational_or_decimal)
+def add_or_sub(value, sample_args, context=None):
+  """Module for adding or subtracting two values."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  is_addition = random.choice([False, True])
+  entropy, sample_args = sample_args.peel()
+
+  if value is None:
+    entropy_p, entropy_q = _entropy_for_pair(entropy)
+    p = number.integer_or_decimal(entropy_p, signed=True)
+    q = number.integer_or_decimal(entropy_q, signed=True)
+  else:
+    entropy = max(entropy, number.entropy_of_value(value))
+    sampler = _value_sampler(value)
+    p = sampler(entropy)
+    if is_addition:
+      q = value - p
+      # Maybe swap for symmetry.
+      if random.choice([False, True]):
+        p, q = q, p
+    else:
+      q = p - value
+      # Maybe swap for symmetry.
+      if random.choice([False, True]):
+        p, q = -q, -p
+
+  p, q = context.sample(sample_args, [p, q])
+
+  if is_addition:
+    return _add_question_or_entity(context, p, q, is_question)
+  else:
+    return _sub_question_or_entity(context, p, q, is_question)
+
+
+def add_or_sub_in_base(sample_args):
+  """Module for addition and subtraction in another base."""
+  context = composition.Context()
+  entropy, sample_args = sample_args.peel()
+  entropy_p, entropy_q = _entropy_for_pair(entropy)
+  p = number.integer(entropy_p, signed=True)
+  q = number.integer(entropy_q, signed=True)
+  base = random.randint(2, 16)
+  if random.choice([False, True]):
+    answer = p + q
+    template = 'Em base {base}, quanto é {p} + {q}?'
+  else:
+    answer = p - q
+    template = 'Em base {base}, quanto é {p} - {q}?'
+  return example.Problem(
+      question=example.question(
+          context,
+          template,
+          base=base,
+          p=display.NumberInBase(p, base),
+          q=display.NumberInBase(q, base)),
+      answer=display.NumberInBase(answer, base))
+
+
+def mul(value, sample_args, context=None):
+  """Returns random question for multiplying two numbers."""
+  del value  # unused
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+  entropy, sample_args = sample_args.peel()
+  entropy_p, entropy_q = _entropy_for_pair(entropy)
+  p = number.integer_or_decimal(entropy_p, True)
+  q = number.integer_or_decimal(entropy_q, True)
+  p, q = context.sample(sample_args, [p, q])
+  answer = p.value * q.value
+
+  if is_question:
+    templates = [
+        '{p}' + ops.MUL_SYMBOL + '{q}',
+        '{p} ' + ops.MUL_SYMBOL + ' {q}',
+        'Calcula {p} * {q}.',
+        'Determina {p} * {q}.',
+        'Multiplica {p} e {q}.',
+        'Produto de {p} e {q}.',
+        'Qual é o produto de {p} e {q}?',
+        '{p} vezes {q}',
+        'Quanto é {p} vezes {q}?',
+    ]
+    template = random.choice(templates)
+    return example.Problem(
+        question=example.question(context, template, p=p, q=q),
+        answer=answer
+    )
+  else:
+    return composition.Entity(
+        context=context,
+        value=answer,
+        description='Seja {self} = {p} * {q}.',
+        p=p, q=q)
+
+
+def div(value, sample_args, context=None):
+  """Returns random question for dividing two numbers."""
+  del value  # unused
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  entropy_1, entropy_q = _entropy_for_pair(entropy)
+
+  q = number.integer(entropy_q, True, min_abs=1)
+
+  if random.choice([False, True]):
+    # Pick p/q with nice integer result.
+    answer = number.integer(entropy_1, True)
+    p = answer * q
+  else:
+    p = number.integer(entropy_1, True)
+    answer = p / q
+
+  p, q = context.sample(sample_args, [p, q])
+
+  if is_question:
+    template = random.choice([
+        'Divide {p} por {q}.',
+        '{p} dividido por {q}',
+        'Quanto é {p} dividido por {q}?',
+        'Calcula {p} dividido por {q}.',
+    ])
+    return example.Problem(
+        question=example.question(context, template, p=p, q=q),
+        answer=answer
+    )
+  else:
+    return composition.Entity(
+        context=context,
+        value=answer,
+        description='Seja {self} ser {p} dividido por {q}.',
+        p=p, q=q)
+
+
+def nearest_integer_root(sample_args):
+  """E.g., "Calculate the cube root of 35 to the nearest integer."."""
+  context = composition.Context()
+
+  # With at least 50% probability, pick square or cube root (these are most
+  # important roots!).
+  if random.choice([False, True]):
+    one_over_exponent = random.randint(2, 3)
+  else:
+    one_over_exponent = random.randint(2, 10)
+
+  entropy, sample_args = sample_args.peel()
+  value = number.integer(entropy, signed=False)
+  answer = int(round(value ** (1 / one_over_exponent)))
+
+  templates = [
+      'Quanto é {value} elevado a 1/{one_over_exponent}, arredondado para o '
+      'inteiro mais próximo?',
+  ]
+
+  if one_over_exponent != 2:  # "What is the second root of 4?" never used.
+    ordinal = str()
+    templates += [
+        'Qual é a {ordinal} raíz de {value} arredondada para o inteiro mais próximo?',
+    ]
+
+  if one_over_exponent == 2:
+    templates += [
+        'Qual é a raíz quadrada de {value} arredondada para o inteiro mais próximo?',
+    ]
+  elif one_over_exponent == 3:
+    templates += [
+        'Qual é a raíz cúbica de {value} arredondada para o inteiro mais próximo?',
+    ]
+
+  template = random.choice(templates)
+
+  ordinal = display.StringOrdinal(one_over_exponent)
+  return example.Problem(
+      question=example.question(
+          context, template, value=value, ordinal=ordinal,
+          one_over_exponent=one_over_exponent),
+      answer=answer)
+
+
+def _calculate(value, sample_args, context, add_sub, mul_div, length=None):
+  """Questions for evaluating arithmetic expressions."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  if value in [_INT, _INT_OR_RATIONAL]:
+    value_entropy = max(1.0, entropy / 4)
+    entropy = max(1.0, entropy - value_entropy)
+    sampler = _value_sampler(value)
+    value = sampler(value_entropy)
+
+  op = arithmetic.arithmetic(
+      value=value, entropy=entropy, add_sub=add_sub, mul_div=mul_div,
+      length=length)
+  context.sample_by_replacing_constants(sample_args, op)
+
+  if is_question:
+    template = random.choice([
+        '{op}',
+        'Quanto é {op}?',
+        'Avalia {op}.',
+        'Calcula {op}.',
+        'Qual é o valor de {op}?',
+    ])
+    return example.Problem(
+        question=example.question(context, template, op=op),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        expression=op,
+        description='Assume que {self} é {op}.',
+        op=op)
+
+
+def add_sub_multiple(value, sample_args, length=None):
+  return _calculate(
+      value, sample_args, None, add_sub=True, mul_div=False, length=length)
+
+
+def mul_div_multiple(value, sample_args, length=None):
+  return _calculate(
+      value, sample_args, None, add_sub=False, mul_div=True, length=length)
+
+
+@composition.module(number.is_integer_or_rational)
+def mixed(value, sample_args, context=None, length=None):
+  return _calculate(
+      value, sample_args, context, add_sub=True, mul_div=True, length=length)
+
+
+def _surd_coefficients(sympy_exp):
+  """Extracts coefficients a, b, where sympy_exp = a + b * sqrt(base)."""
+  sympy_exp = sympy.simplify(sympy.expand(sympy_exp))
+
+  def extract_b(b_sqrt_base):
+    """Returns b from expression of form b * sqrt(base)."""
+    if isinstance(b_sqrt_base, sympy.Pow):
+      # Just form sqrt(base)
+      return 1
+    else:
+      assert isinstance(b_sqrt_base, sympy.Mul)
+      assert len(b_sqrt_base.args) == 2
+      assert b_sqrt_base.args[0].is_rational
+      assert isinstance(b_sqrt_base.args[1], sympy.Pow)  # should be sqrt.
+      return b_sqrt_base.args[0]
+
+  if sympy_exp.is_rational:
+    # Form: a.
+    return sympy_exp, 0
+  elif isinstance(sympy_exp, sympy.Add):
+    # Form: a + b * sqrt(base)
+    assert len(sympy_exp.args) == 2
+    assert sympy_exp.args[0].is_rational
+    a = sympy_exp.args[0]
+    b = extract_b(sympy_exp.args[1])
+    return a, b
+  else:
+    # Form: b * sqrt(base).
+    return 0, extract_b(sympy_exp)
+
+
+def _surd_split_entropy_two(entropy):
+  entropy_left = entropy / 2
+  if entropy_left < 1:
+    entropy_left = 0
+  entropy_right = entropy - entropy_left
+  if random.choice([False, True]):
+    entropy_left, entropy_right = entropy_right, entropy_left
+  return entropy_left, entropy_right
+
+
+def _sample_surd(base, entropy, max_power, multiples_only):
+  """An expression that can be reduced to a + b * sqrt(base).
+
+  For example, if base=3, then the following are valid expressions:
+
+  *   sqrt(12)   (reduces to 2 * sqrt(3))
+  *   sqrt(3) - 10 * sqrt(3)  (reduces to -9 * sqrt(3))
+  *   sqrt(15) / sqrt(5)  (reduces to sqrt(3)).
+  *   4 * sqrt(3) / 2
+  *   2 + sqrt(3)
+  *   1 / (1 + sqrt(3))  (reduces to -1/2 + (-1/2) sqrt(3))
+
+  However, 1 + 2 * sqrt(3) is not valid, as it does not reduce to the form
+  a * sqrt(3).
+
+  Args:
+    base: The value inside the square root.
+    entropy: Float >= 0; used for randomness.
+    max_power: Integer >= 1; the max power used in expressions. If 1 then
+        disables.
+    multiples_only: Whether the surd should be an integer multiple of
+        sqrt(base).
+
+  Returns:
+    Instance of `ops.Op`.
+  """
+  if entropy <= 0:
+    return ops.Sqrt(base)
+
+  def add_or_sub_():
+    # Add or subtract two such types.
+    entropy_left, entropy_right = _surd_split_entropy_two(entropy)
+    left = _sample_surd(base, entropy_left, max_power, multiples_only)
+    right = _sample_surd(base, entropy_right, max_power, multiples_only)
+    op = random.choice([ops.Add, ops.Sub])
+    return op(left, right)
+
+  def mul_by_integer():
+    entropy_k = min(1, entropy)
+    left = number.integer(entropy_k, signed=True, min_abs=1)
+    right = _sample_surd(base, entropy - entropy_k, max_power, multiples_only)
+    if random.choice([False, True]):
+      left, right = right, left
+    return ops.Mul(left, right)
+
+  def div_by_sqrt_k():
+    """Do sqrt(k * base) / sqrt(k)."""
+    entropy_k = min(1, entropy)
+    k = number.integer(entropy_k, signed=False, min_abs=2)
+    entropy_left, entropy_right = _surd_split_entropy_two(entropy - entropy_k)
+    k_base_expr = _sample_surd(k * base, entropy_left, max_power, True)
+    while True:
+      k_expr = _sample_surd(k, entropy_right, max_power, True)
+      if k_expr.sympy() != 0:
+        break
+    return ops.Div(k_base_expr, k_expr)
+
+  def square_k():
+    """Do sqrt(k * k * base)."""
+    entropy_k = min(1, entropy)
+    k = number.integer(entropy_k, signed=False, min_abs=2)
+    return _sample_surd(
+        k * k * base, entropy - entropy_k, max_power, multiples_only)
+
+  def surd_plus_integer():
+    """Do surd + integer."""
+    entropy_k = min(1, entropy)
+    left = number.integer(entropy_k, signed=True)
+    assert not multiples_only
+    right = _sample_surd(base, entropy - entropy_k, max_power, False)
+    if random.choice([True, False]):
+      left, right = right, left
+    return ops.Add(left, right)
+
+  def power():
+    """Do surd**2."""
+    assert not multiples_only
+    surd = _sample_surd(base, entropy, max_power=1, multiples_only=False)
+    return ops.Pow(surd, 2)
+
+  choices = [add_or_sub_, mul_by_integer]
+  if not multiples_only:
+    choices += [surd_plus_integer]
+    if max_power > 1:
+      choices += [power]
+  if base < 64:  # prevent value inside sqrt from getting too big
+    choices += [div_by_sqrt_k, square_k]
+  which = random.choice(choices)
+  return which()
+
+
+def simplify_surd(value, sample_args, context=None):
+  """E.g., "Simplify (2 + 5*sqrt(3))**2."."""
+  del value  # unused
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  while True:
+    base = random.randint(2, 20)
+    if sympy.Integer(base).is_prime:
+      break
+  num_primes_less_than_20 = 8
+  entropy -= math.log10(num_primes_less_than_20)
+  exp = _sample_surd(base, entropy, max_power=2, multiples_only=False)
+  simplified = sympy.expand(sympy.simplify(exp))
+
+  template = random.choice([
+      'Simplificar {exp}.',
+  ])
+  return example.Problem(
+      question=example.question(context, template, exp=exp),
+      answer=simplified)
diff --git a/mathematics_dataset/mathematics_dataset/modules/arithmetic_test.py b/mathematics_dataset/mathematics_dataset/modules/arithmetic_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..0be9ff7b54bfae52514f4160257e7f3fbf9428c0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/arithmetic_test.py
@@ -0,0 +1,64 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.modules.arithmetic."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from modules import arithmetic
+import sympy
+
+
+class ArithmeticTest(absltest.TestCase):
+
+  def testSurdCoefficients(self):
+    exp = sympy.sympify('1')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (1, 0))
+
+    exp = sympy.sympify('1/2')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (1/2, 0))
+
+    exp = sympy.sympify('sqrt(2)')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (0, 1))
+
+    exp = sympy.sympify('3*sqrt(2)')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (0, 3))
+
+    exp = sympy.sympify('3*sqrt(5)/2')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (0, 3/2))
+
+    exp = sympy.sympify('1 + 3 * sqrt(2)')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (1, 3))
+
+    exp = sympy.sympify('1/2 + 3 * sqrt(5) / 2')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (1/2, 3/2))
+
+    exp = sympy.sympify('sqrt(2)/(-1 + 2*sqrt(2))**2')
+    self.assertEqual(arithmetic._surd_coefficients(exp),
+                     (8/49, 9/49))
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/modules/calculus.py b/mathematics_dataset/mathematics_dataset/modules/calculus.py
new file mode 100644
index 0000000000000000000000000000000000000000..46e6c9948f86d5fdf38e6ef18424ea6189496654
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/calculus.py
@@ -0,0 +1,198 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Calculus related questions, e.g., "differentiate x**2"."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import math
+import random
+
+# Dependency imports
+import example
+from sample import polynomials
+from util import composition
+from util import display
+import numpy as np
+from six.moves import range
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+
+
+def _make_modules(entropy):
+  """Returns modules given "difficulty" parameters."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  sample_args_composed = composition.PreSampleArgs(2, 4, *entropy)
+
+  return {
+      'differentiate_composed': functools.partial(
+          differentiate_univariate, None, sample_args_composed),
+      'differentiate': functools.partial(differentiate, None, sample_args_pure),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(entropy_fn(_ENTROPY_TRAIN))
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(_ENTROPY_INTERPOLATE)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  return {
+  }
+
+
+def _generate_polynomial(num_variables, entropy, derivative_order,
+                         derivative_axis):
+  """Returns polynomial."""
+  # Note: numpy randint has upper bound as ) not ], unlike python random.randint
+  degrees = np.random.randint(1, 4, [num_variables])
+  degrees[derivative_axis] = np.random.randint(0, 4)  # allow to be zero here.
+
+  coefficients = polynomials.sample_coefficients(degrees, entropy)
+
+  # We also generate coefficients that will disappear when differentiated.
+  # Thus we don't account for the entropy used here.
+  assert derivative_order > 0
+  degrees[derivative_axis] = derivative_order - 1
+  extra_coefficients = polynomials.sample_coefficients(degrees, entropy)
+
+  return np.concatenate(
+      [extra_coefficients, coefficients], axis=derivative_axis)
+
+
+def _template(module_count, derivative_order, num_variables):
+  """Selects appropriate template."""
+  templates = [
+      'Encontra a {nth} derivada de {eq} em relação a {var}.',
+      'Qual é a {nth} derivada de {eq} em relação a {var}?',
+  ]
+  if derivative_order == 1:
+    templates += [
+        'Deriva {eq} em relação a {var}.',
+        'Diferencia {eq} em relação a {var}.',
+        'Qual é a derivada de {eq} em relação a {var}?',
+    ]
+
+  derivative_variable_is_unambiguous = num_variables == 1 and module_count == 1
+  if derivative_variable_is_unambiguous:
+    templates += [
+        'Encontra a {nth} derivada de {eq}.',
+        'Qual é a {nth} derivada de {eq}?',
+    ]
+    if derivative_order == 1:
+      templates += [
+          'Deriva {eq}.',
+          'Qual é a derivada de {eq}?',
+      ]
+
+  return random.choice(templates)
+
+
+def _sample_integrand(coefficients, derivative_order, derivative_axis, entropy):
+  """Integrates `coefficients` and adds sampled "constant" terms."""
+  coefficients = np.asarray(coefficients)
+
+  # Integrate (with zero for constant terms).
+  integrand = coefficients
+  for _ in range(derivative_order):
+    integrand = polynomials.integrate(integrand, derivative_axis)
+
+  # Add on sampled constant terms.
+  constant_degrees = np.array(integrand.shape) - 1
+  constant_degrees[derivative_axis] = derivative_order - 1
+  extra_coeffs = polynomials.sample_coefficients(constant_degrees, entropy)
+  pad_amount = coefficients.shape[derivative_axis]
+  pad = [(0, pad_amount if i == derivative_axis else 0)
+         for i in range(coefficients.ndim)]
+  extra_coeffs = np.pad(extra_coeffs, pad, 'constant', constant_values=0)
+  return integrand + extra_coeffs
+
+
+def _differentiate_polynomial(value, sample_args, context, num_variables):
+  """Generates a question for differentiating a polynomial."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  if value is not None:
+    num_variables = value.coefficients.ndim
+
+  entropy, sample_args = sample_args.peel()
+  max_derivative_order = 3
+  derivative_order = random.randint(1, max_derivative_order)
+  entropy = max(0, entropy - math.log10(max_derivative_order))
+
+  derivative_axis = random.randint(0, num_variables - 1)
+  if value is None:
+    coefficients = _generate_polynomial(
+        num_variables, entropy, derivative_order, derivative_axis)
+  else:
+    coefficients = _sample_integrand(
+        value.coefficients, derivative_order, derivative_axis, entropy)
+
+  (entity,) = context.sample(
+      sample_args, [composition.Polynomial(coefficients)])
+
+  value = coefficients
+  for _ in range(derivative_order):
+    value = polynomials.differentiate(value, axis=derivative_axis)
+  nth = display.StringOrdinal(derivative_order)
+
+  if entity.has_expression():
+    polynomial = entity.expression
+    variables = entity.polynomial_variables
+  else:
+    variables = [sympy.Symbol(context.pop()) for _ in range(num_variables)]
+    polynomial = entity.handle.apply(*variables)
+  variable = variables[derivative_axis]
+
+  if is_question:
+    template = _template(context.module_count, derivative_order, len(variables))
+    answer = polynomials.coefficients_to_polynomial(value, variables).sympy()
+    return example.Problem(
+        question=example.question(
+            context, template, eq=polynomial, var=variable, nth=nth),
+        answer=answer)
+  else:
+    fn_symbol = context.pop()
+    variables_string = ', '.join(str(variable) for variable in variables)
+    assert len(variables) == 1  # since below we don't specify var we diff wrt
+    return composition.Entity(
+        context=context,
+        value=composition.Polynomial(value),
+        description='Seja {fn}({variables}) a {nth} derivada de {eq}.',
+        handle=composition.FunctionHandle(fn_symbol),
+        fn=fn_symbol, variables=variables_string, nth=nth, eq=polynomial)
+
+
+def differentiate_univariate(value, sample_args, context=None):
+  return _differentiate_polynomial(value, sample_args, context, 1)
+
+
+@composition.module(composition.is_polynomial)
+def differentiate(value, sample_args, context=None):
+  num_variables = random.randint(1, 4)
+  return _differentiate_polynomial(value, sample_args, context, num_variables)
diff --git a/mathematics_dataset/mathematics_dataset/modules/calculus_test.py b/mathematics_dataset/mathematics_dataset/modules/calculus_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..c7693d1f6f498ed34e450b58eebb323fea254e5d
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/calculus_test.py
@@ -0,0 +1,43 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.modules.calculus."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from modules import calculus
+import tensorflow as tf
+
+
+class CalculusTest(tf.test.TestCase):
+
+  def testSampleIntegrand(self):
+    # y + 2*x + 3*x**2
+    coefficients = [[0, 1], [2, 0], [3, 0]]
+    derivative_order = 1
+    derivative_axis = 0
+    # const + x*y + x**2 + x**3
+    expected = [[0, 1], [1, 0], [1, 0]]
+    entropy = 4
+    result = calculus._sample_integrand(
+        coefficients, derivative_order, derivative_axis, entropy)
+    result = result[1:, :]  # ignore random constant terms
+    self.assertAllEqual(result, expected)
+
+
+if __name__ == '__main__':
+  tf.test.main()
diff --git a/mathematics_dataset/mathematics_dataset/modules/comparison.py b/mathematics_dataset/mathematics_dataset/modules/comparison.py
new file mode 100644
index 0000000000000000000000000000000000000000..c88ce8c8c9d76d97bf134479e46d13eca9f82fb7
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/comparison.py
@@ -0,0 +1,455 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Comparisons, e.g. "is 2 > 3?"."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import random
+
+# Dependency imports
+import example
+from sample import number
+from sample import ops
+from util import composition
+from util import display
+import numpy as np
+from six.moves import range
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+_ENTROPY_EXTRAPOLATE = (12, 12)
+
+_EXTRAPOLATION_EXTRA_COUNT = 2
+
+_PROB_EQUAL = 0.2
+
+
+def _make_modules(entropy):
+  """Returns modules given "difficulty" parameters."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  sample_args_composed = composition.PreSampleArgs(2, 4, *entropy)
+
+  return {
+      'pair': functools.partial(pair, sample_args_pure),
+      'pair_composed': functools.partial(pair, sample_args_composed),
+      'kth_biggest': functools.partial(kth_biggest, sample_args_pure),
+      'kth_biggest_composed': functools.partial(
+          kth_biggest, sample_args_composed),
+      'closest': functools.partial(closest, sample_args_pure),
+      'closest_composed': functools.partial(closest, sample_args_composed),
+      'sort': functools.partial(sort, sample_args_pure),
+      'sort_composed': functools.partial(sort, sample_args_composed),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(entropy_fn(_ENTROPY_TRAIN))
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(_ENTROPY_INTERPOLATE)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *_ENTROPY_EXTRAPOLATE)
+
+  def sort_count():
+    lower = _sort_count_range(_ENTROPY_TRAIN[1])[1]
+    return random.randint(lower + 1, lower + _EXTRAPOLATION_EXTRA_COUNT)
+  def closest_count():
+    lower = _closest_count_range(_ENTROPY_TRAIN[1])[1]
+    return random.randint(lower + 1, lower + _EXTRAPOLATION_EXTRA_COUNT)
+  def kth_biggest_more():
+    return kth_biggest(sample_args_pure, count=sort_count())
+  def sort_more():
+    return sort(sample_args_pure, count=sort_count())
+  def closest_more():
+    return closest(sample_args_pure, count=closest_count())
+
+  return {
+      'kth_biggest_more': kth_biggest_more,
+      'sort_more': sort_more,
+      'closest_more': closest_more,
+  }
+
+
+def _make_comparison_question(context, left, right):
+  """Makes a question for comparing two values."""
+  if random.choice([False, True]) and sympy.Ne(left.value, right.value):
+    # Do question of form: "Which is bigger: a or b?".
+    if random.choice([False, True]):
+      answer = (
+          left.handle if sympy.Gt(left.value, right.value) else right.handle)
+      template = random.choice([
+          'Qual é maior: {left} ou {right}?',
+          'Qual é superior: {left} ou {right}?',
+      ])
+    else:
+      answer = (
+          left.handle if sympy.Lt(left.value, right.value) else right.handle)
+      template = random.choice([
+          'Qual é menor: {left} ou {right}?',
+      ])
+    return example.Problem(
+        question=example.question(context, template, left=left, right=right),
+        answer=answer)
+
+  comparisons = {
+      '<': sympy.Lt,
+      '<=': sympy.Le,
+      '>': sympy.Gt,
+      '>=': sympy.Ge,
+      '=': sympy.Eq,
+      '!=': sympy.Ne,
+  }
+
+  templates = {
+      '<': [
+          'Será {left} ' + ops.LT_SYMBOL + ' {right}?',
+          'Será {left} menor que {right}?',
+          'Será {left} menor do que {right}?',
+      ],
+      '<=': [
+          'Será {left} ' + ops.LE_SYMBOL + ' {right}?',
+          'Será {left} menor ou igual a {right}?',
+          'Será {left}, no máximo, {right}?',
+          'Será {left}, no máximo, tão grande quanto {right}?',
+      ],
+      '>': [
+          'Será {left} ' + ops.GT_SYMBOL + ' {right}?',
+          'Será {left} maior que {right}?',
+          'Será {left} maior do que {right}?',
+      ],
+      '>=': [
+          'Será {left} ' + ops.GE_SYMBOL + ' {right}?',
+          'É verdade que {left} é maior ou igual a {right}?',
+          'É verdade que {left} é, no mínimo, {right}?',
+          'É verdade que {left} é, pelo menos, maior do que {right}?',
+      ],
+      '=': [
+          'Será que {left} ' + ops.EQ_SYMBOL + ' {right}?',
+          'Serão {left} e {right} iguais?',
+          'Será {left} igual a {right}?',
+          'Será que {left} e {right} possuem o mesmo valor?',
+      ],
+      '!=': [
+          'Será que {left} ' + ops.NE_SYMBOL + ' {right}?',
+          'Será {left} diferente de {right}?',
+          'Será que {left} e {right} são desiguais?',
+          'Será que {left} e {right} não são equivalentes?',
+          'Será que {left} e {right} não são iguais?',
+          'Será que {left} e {right} possuem valores diferentes?',
+      ],
+  }
+
+  comparison = random.choice(list(comparisons.keys()))
+  template = random.choice(templates[comparison])
+  question = example.question(context, template, left=left, right=right)
+  answer = comparisons[comparison](left.value, right.value)
+
+  return example.Problem(question=question, answer=answer)
+
+
+def integer_or_rational_or_decimal(entropy):
+  if random.choice([False, True]):
+    return number.integer_or_decimal(entropy, signed=True)
+  else:
+    return number.integer_or_rational(entropy, signed=True)
+
+
+def pair(sample_args, context=None):
+  """Compares two numbers, e.g., "is 1/2 < 0.5?"."""
+  if context is None:
+    context = composition.Context()
+  entropy, sample_args = sample_args.peel()
+
+  def integers_close():
+    entropy_diff, entropy_left = entropy * np.random.dirichlet([1, 3])
+    left = number.integer(entropy_left, True)
+    right = left + number.integer(entropy_diff, True)
+    return left, right
+
+  def rational_and_integer():
+    # Pick rational, and integer close to rational evaluation
+    left = number.non_integer_rational(entropy, True)
+    right = int(round(left)) + random.randint(-1, 1)
+    return left, right
+
+  def independent():
+    # Return an independent pair.
+    entropy_left, entropy_right = entropy * np.random.dirichlet([1, 1])
+    left = integer_or_rational_or_decimal(entropy_left)
+    right = integer_or_rational_or_decimal(entropy_right)
+    return left, right
+
+  generator = random.choice([integers_close, rational_and_integer, independent])
+
+  left, right = generator()
+
+  # maybe swap for symmetry
+  if random.choice([False, True]):
+    left, right = right, left
+  left, right = context.sample(sample_args, [left, right])
+
+  return _make_comparison_question(context, left, right)
+
+
+def _entities_to_list(entities):
+  entity_dict = {}
+  values_template = ''
+  for i, entity in enumerate(entities):
+    if i > 0:
+      values_template += ', '
+    entity_name = 'entity_{}'.format(i)
+    entity_dict[entity_name] = entity
+    values_template += '{' + entity_name + '}'
+  return entity_dict, values_template
+
+
+def _entities_to_choices(entities, answer):
+  """Generate a multichoice question template."""
+  if len(entities) > 26:
+    raise ValueError('Too many choices: {}'.format(len(entities)))
+
+  entity_dict = {}
+  choices_template = ''
+  answer_choice = None
+  for i, entity in enumerate(entities):
+    choices_template += '  '
+    entity_name = 'entity_{}'.format(i)
+    entity_dict[entity_name] = entity
+    letter = chr(ord('a') + i)
+    choices_template += '({letter}) {{{entity_name}}}'.format(
+        letter=letter, entity_name=entity_name)
+    if entity is answer:
+      assert answer_choice is None
+      answer_choice = letter
+
+  assert answer_choice is not None
+  return entity_dict, choices_template, answer_choice
+
+
+def _mark_choice_letters_used(count, context):
+  """Marks the choice letters as used."""
+  for i in range(count):
+    context.mark_used(chr(ord('a') + i))
+
+
+def _kth_biggest_list_question(context, entities, adjective, answer):
+  """Ask for the biggest (or smallest, or second biggest, etc) in a list."""
+  entity_dict, values_template = _entities_to_list(entities)
+
+  question = example.question(
+      context, 'Qual é o {adjective} valor em ' + values_template + '?',
+      adjective=adjective, **entity_dict)
+  return example.Problem(question=question, answer=answer.handle)
+
+
+def _kth_biggest_multichoice_question(context, entities, adjective, answer):
+  """Ask for the biggest (or smallest, or second biggest, etc) of choices."""
+  entity_dict, choices_template, answer_choice = _entities_to_choices(
+      entities, answer)
+  question = example.question(
+      context, 'Qual valor é {adjective}?' + choices_template,
+      adjective=adjective, **entity_dict)
+  return example.Problem(question=question, answer=answer_choice)
+
+
+def _entity_sort_key(entity):
+  return sympy.default_sort_key(entity.value)
+
+
+def _sort_count_range(entropy):
+  min_ = 3
+  return min_, min_ + int(entropy/2)
+
+
+def _unique_values(entropy, only_integers=False, count=None):
+  """Generates unique values."""
+  if count is None:
+    count = random.randint(*_sort_count_range(entropy))
+
+  if only_integers:
+    sampler = functools.partial(number.integer, signed=True)
+  else:
+    sampler = integer_or_rational_or_decimal
+
+  for _ in range(1000):
+    entropies = entropy * np.random.dirichlet(np.ones(count))
+    entropies = np.maximum(1, entropies)
+    values = [sampler(ent) for ent in entropies]
+    if len(sympy.FiniteSet(*values)) == len(values):
+      return values
+  raise ValueError('Could not generate {} unique values with entropy={}'
+                   .format(count, entropy))
+
+
+def kth_biggest(sample_args, count=None):
+  """Asks for the kth biggest value in a list."""
+  sample_args = sample_args()
+  context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  values = _unique_values(entropy, count=count)
+  count = len(values)
+
+  display_multichoice = random.choice([False, True])
+  if display_multichoice:
+    _mark_choice_letters_used(count, context)
+
+  entities = context.sample(sample_args, values)
+  sorted_entities = sorted(entities, key=_entity_sort_key)
+  ordinal = random.randint(1, count)
+
+  if random.choice([False, True]):
+    # Do from biggest.
+    answer = sorted_entities[-ordinal]
+    adjective = 'maior'
+  else:
+    # Do from smallest.
+    answer = sorted_entities[ordinal - 1]
+    adjective = 'menor'
+
+  if ordinal > 1:
+    ordinal_PT_masc = ""
+    for palavra in str(display.StringOrdinal(ordinal)).split():
+      ordinal_PT_masc += palavra[:-1] + "o"
+    adjective = ordinal_PT_masc + ' ' + adjective
+
+  if display_multichoice:
+    return _kth_biggest_multichoice_question(
+        context=context, entities=entities, adjective=adjective, answer=answer)
+  else:
+    return _kth_biggest_list_question(
+        context=context, entities=entities, adjective=adjective, answer=answer)
+
+
+def _closest_in_list_question(context, entities, target, adjective, answer):
+  """Ask for the closest to a given value in a list."""
+  entity_dict, values_template = _entities_to_list(entities)
+
+  question = example.question(
+      context,
+      'Qual valor está {adjective} de {target} em ' + values_template + '?',
+      adjective=adjective, target=target, **entity_dict)
+  return example.Problem(question=question, answer=answer.handle)
+
+
+def _closest_multichoice_question(context, entities, target, adjective, answer):
+  """Ask for the closest to a given value in a set of choices."""
+  entity_dict, choices_template, answer_choice = _entities_to_choices(
+      entities, answer)
+
+  question = example.question(
+      context,
+      'Qual está {adjective} de {target}?' + choices_template,
+      adjective=adjective, target=target, **entity_dict)
+  return example.Problem(question=question, answer=answer_choice)
+
+
+def _closest_count_range(entropy):
+  min_ = 3
+  return min_, min_ + int(entropy/3)
+
+
+def closest(sample_args, count=None):
+  """Ask for the closest to a given value in a list."""
+  sample_args = sample_args()
+  context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  if count is None:
+    count = random.randint(*_closest_count_range(entropy))
+
+  display_multichoice = random.choice([False, True])
+  if display_multichoice:
+    _mark_choice_letters_used(count, context)
+
+  entropy_target, entropy_list = entropy * np.random.dirichlet([1, count])
+  target = integer_or_rational_or_decimal(entropy_target)
+
+  while True:
+    value_entropies = entropy_list * np.random.dirichlet(np.ones(count))
+    value_entropies = np.maximum(1, value_entropies)
+    values = [integer_or_rational_or_decimal(ent) for ent in value_entropies]
+    differences = [abs(sympy.sympify(value) - target) for value in values]
+    if len(sympy.FiniteSet(*differences)) == count:  # all differences unique
+      break
+
+  target_and_entities = context.sample(sample_args, [target] + values)
+  target = target_and_entities[0]
+  entities = target_and_entities[1:]
+
+  min_difference = min(differences)
+  answer_index = differences.index(min_difference)
+  answer = entities[answer_index]
+  adjective = random.choice(['mais próximo', 'mais perto'])
+
+  if display_multichoice:
+    return _closest_multichoice_question(
+        context=context, entities=entities, target=target, adjective=adjective,
+        answer=answer)
+  else:
+    return _closest_in_list_question(
+        context=context, entities=entities, target=target, adjective=adjective,
+        answer=answer)
+
+
+def sort(sample_args, count=None):
+  """Ask to sort numbers in increasing or decreasing order."""
+  sample_args = sample_args()
+  context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  # Sometimes just integers, to allow for more terms in a short space.
+  values = _unique_values(
+      entropy, only_integers=random.choice([False, True]), count=count)
+
+  entities = context.sample(sample_args, values)
+
+  unsorted_dict, unsorted_template = _entities_to_list(entities)
+
+  ascending = random.choice([False, True])
+  templates = [
+      'Ordena ' + unsorted_template + ' por ordem {direction}.',
+      'Coloca ' + unsorted_template + ' em ordem {direction}.',
+  ]
+  if ascending:
+    templates.append('Ordena ' + unsorted_template + '.')
+    direction = random.choice(['ascendente', 'crescente'])
+  else:
+    direction = random.choice(['descendente', 'decrescente'])
+  template = random.choice(templates)
+
+  sorted_entities = sorted(
+      entities, key=_entity_sort_key, reverse=(not ascending))
+  answer = ''
+  for i, entity in enumerate(sorted_entities):
+    if i > 0:
+      answer += ', '
+    answer += str(entity.handle)
+
+  return example.Problem(
+      question=example.question(
+          context, template, direction=direction, **unsorted_dict),
+      answer=answer)
diff --git a/mathematics_dataset/mathematics_dataset/modules/measurement.py b/mathematics_dataset/mathematics_dataset/modules/measurement.py
new file mode 100644
index 0000000000000000000000000000000000000000..08a1a45d6e512346f162fbc7f87879b6bd7042e0
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/measurement.py
@@ -0,0 +1,279 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Measurement questions, e.g., "How many hours are there in a day?"."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+import functools
+import random
+
+# Dependency imports
+import example
+from modules import train_test_split
+from sample import number
+from util import composition
+from util import display
+import six
+import sympy
+
+
+def _make_modules(is_train):
+  """Returns modules, with split based on the boolean `is_train`."""
+  return {
+      'conversion': functools.partial(
+          conversion, is_train=is_train, is_extrapolation=False),
+      'time': functools.partial(time, is_train=is_train),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  del entropy_fn  # unused
+  return _make_modules(is_train=True)
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(is_train=False)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  return {
+      'conversion': functools.partial(
+          conversion, is_train=False, is_extrapolation=True),
+  }
+
+
+Unit = collections.namedtuple('Unit', ('name', 'symbol'))
+
+
+MICRO_SYMBOL = 'u'
+
+
+LENGTH = {
+    Unit('metro', 'm'): 1,
+    Unit('quilómetro', 'km'): 1000,
+    Unit('centímetro', 'cm'): sympy.Rational(1, 100),
+    Unit('millímetro', 'mm'): sympy.Rational(1, 1000),
+    Unit('micrómetro', 'um'): sympy.Rational(1, 1e6),
+    Unit('nanómetro', 'nm'): sympy.Rational(1, 1e9),
+}
+
+TIME = {
+    Unit('segundo', 's'): 1,
+    Unit('minuto', None): 60,
+    Unit('hora', None): 60*60,
+    Unit('dia', None): 24*60*60,
+    Unit('semana', None): 7*24*60*60,
+    Unit('miliseguno', 'ms'): sympy.Rational(1, 1e3),
+    Unit('microsegundo', MICRO_SYMBOL + 's'): sympy.Rational(1, 1e6),
+    Unit('nanosegundo', 'ns'): sympy.Rational(1, 1e9),
+}
+
+TIME_YEARLY = {
+    Unit('ano', None): 1,
+    Unit('década', None): 10,
+    Unit('século', None): 100,
+    Unit('milénio', None): 1000,
+    Unit('mês', None): sympy.Rational(1, 12),
+}
+
+MASS = {
+    Unit('quilograma', 'kg'): 1,  # Yes, the *kilo*gram is the SI base unit.
+    Unit('tonelada', 't'): 1000,
+    Unit('grama', 'g'): sympy.Rational(1, 1e3),
+    Unit('milligrama', 'mg'): sympy.Rational(1, 1e6),
+    Unit('micrograma', MICRO_SYMBOL + 'g'): sympy.Rational(1, 1e9),
+    Unit('nanograma', 'ng'): sympy.Rational(1, 1e12),
+}
+
+VOLUME = {
+    Unit('litro', 'l'): 1,
+    Unit('mililitro', 'ml'): sympy.Rational(1, 1000),
+}
+
+
+DIMENSIONS = [LENGTH, TIME, TIME_YEARLY, MASS, VOLUME]
+
+
+def pluralize(name):
+  if name == 'century':
+    return 'centuries'
+  if name == 'mês':
+    return 'meses'
+  if name == 'millennium':
+    return 'millennia'
+  return name + 's'
+
+
+def _factor_non_decimal(value):
+  """Extras x dividing value such that x is coprime to 2 and 5."""
+  result = 1
+  factors = sympy.factorint(value)
+  for factor, power in six.iteritems(factors):
+    if factor not in [2, 5]:
+      result *= factor ** power
+  return result
+
+
+def _sample_conversion_decimal(dimension, is_extrapolation):
+  """Samples to and from units and values."""
+  base_unit, target_unit = random.sample(list(dimension.keys()), 2)
+  scale = sympy.Rational(dimension[base_unit]) / dimension[target_unit]
+  scale_non_decimal = _factor_non_decimal(sympy.denom(scale))
+  entropy = 9 if is_extrapolation else 7
+  base_value = number.non_integer_decimal(entropy, signed=False)
+  base_value = display.Decimal(base_value.value * scale_non_decimal)
+  target_value = display.Decimal(base_value.value * scale)
+  return base_value, base_unit, target_value, target_unit
+
+
+def _conversion_decimal(context, is_train, is_extrapolation):
+  """E.g., "How many grams are in 5kg?"."""
+  dimension = random.choice(DIMENSIONS)
+  while True:
+    base_value, base_unit, target_value, target_unit = (
+        _sample_conversion_decimal(dimension, is_extrapolation))
+    if train_test_split.is_train(base_value) == is_train:
+      break
+
+  templates = [
+      'Quantos(as) {target_name} estão em {base_value} {base_name}?',
+      'Quanto é {base_value} {base_name} em {target_name}?',
+      'Converte {base_value} {base_name} para {target_name}.',
+  ]
+  if base_unit.symbol is not None:
+    templates += [
+        'Quantos(as) {target_name} estão em {base_value}{base_symbol}?',
+        'Quanto é {base_value}{base_symbol} em {target_name}?',
+        'Converte {base_value}{base_symbol} para {target_name}.',
+    ]
+  template = random.choice(templates)
+
+  base_name = pluralize(base_unit.name)
+  target_name = pluralize(target_unit.name)
+
+  question = example.question(
+      context,
+      template,
+      base_name=base_name,
+      base_symbol=base_unit.symbol,
+      base_value=base_value,
+      target_name=target_name)
+  return example.Problem(question=question, answer=target_value)
+
+
+def _conversion_fraction(context, is_train):
+  """E.g., "How many grams are in three quarters of a kg?"."""
+  dimension = random.choice(DIMENSIONS)
+
+  # Limit probability of giving zero answer.
+  allow_zero = random.random() < 0.2
+
+  # Repeat until we find a pair with an integral answer. (Avoids ambiguity with
+  # decimals.)
+  while True:
+    base_unit, target_unit = random.sample(list(dimension.keys()), 2)
+    base_value = number.non_integer_rational(2, signed=False)
+    if train_test_split.is_train(base_value) != is_train:
+      continue
+    answer = (base_value * sympy.Rational(dimension[base_unit])
+              / sympy.Rational(dimension[target_unit]))
+    if (abs(answer) <= 100000
+        and sympy.denom(answer) == 1
+        and (allow_zero or answer != 0)):
+      break
+
+  template = random.choice([
+      'Quantos(as) {target_name} estão em {base_value} de um(a) {base_name}?',
+      'Quanto é {base_value} de um(a) {base_name} em um(a) {target_name}?',
+  ])
+
+  if sympy.denom(base_value) > 20 or random.choice([False, True]):
+    base_value_string = base_value  # Will be represented as e.g., 2/3.
+  else:
+    base_value_string = display.StringNumber(base_value)  # e.g., two thirds
+
+  question = example.question(
+      context, template,
+      base_name=base_unit.name,
+      base_value=base_value_string,
+      target_name=pluralize(target_unit.name))
+  return example.Problem(question=question, answer=answer)
+
+
+def conversion(is_train, is_extrapolation):
+  """Conversion question, in decimal or fraction."""
+  context = composition.Context()
+  # TODO(b/124038528): implement extrapolation for fraction conversions too
+  if is_extrapolation or random.choice([False, True]):
+    return _conversion_decimal(
+        context, is_train=is_train, is_extrapolation=is_extrapolation)
+  else:
+    return _conversion_fraction(context, is_train=is_train)
+
+
+def time(is_train):
+  """Questions for calculating start, end, or time differences."""
+  context = composition.Context()
+  start_minutes = random.randint(1, 24*60 - 1)
+  while True:
+    duration_minutes = random.randint(1, 12*60 - 1)
+    if train_test_split.is_train(duration_minutes) == is_train:
+      break
+  end_minutes = start_minutes + duration_minutes
+
+  def format_12hr(minutes):
+    """Format minutes from midnight in 12 hr format."""
+    hours = (minutes // 60) % 24
+    minutes %= 60
+    am_pm = 'AM' if hours < 12 else 'PM'
+    hours = (hours - 1) % 12 + 1
+    return '{}:{:02} {}'.format(hours, minutes, am_pm)
+
+  start = format_12hr(start_minutes)
+  end = format_12hr(end_minutes)
+
+  which_question = random.randint(0, 3)
+  if which_question == 0:
+    # Question: What is start = end - duration?
+    template = random.choice([
+        'O que é {duration} minutos antes de {end}?',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, duration=duration_minutes, end=end),
+        answer=start)
+  elif which_question == 1:
+    # Question: What is end = start + duration?
+    template = random.choice([
+        'O que é {duration} minutos despois de {start}?',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, duration=duration_minutes, start=start),
+        answer=end)
+  else:
+    # Question: What is duration = end - start?
+    template = random.choice([
+        'Quantos minutos existem entre {start} e {end}?',
+    ])
+    return example.Problem(
+        question=example.question(context, template, start=start, end=end),
+        answer=duration_minutes)
diff --git a/mathematics_dataset/mathematics_dataset/modules/modules.py b/mathematics_dataset/mathematics_dataset/modules/modules.py
new file mode 100644
index 0000000000000000000000000000000000000000..a96a6741bb23d80d2f9fe5f512b21712647e29f2
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/modules.py
@@ -0,0 +1,58 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""The various mathematics modules."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from modules import algebra
+from modules import arithmetic
+from modules import calculus
+from modules import comparison
+from modules import measurement
+from modules import numbers
+from modules import polynomials
+from modules import probability
+import six
+
+
+all_ = {
+    'algebra': algebra,
+    'arithmetic': arithmetic,
+    'calculus': calculus,
+    'comparison': comparison,
+    'measurement': measurement,
+    'numbers': numbers,
+    'polynomials': polynomials,
+    'probability': probability,
+}
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return {
+      name: module.train(entropy_fn) for name, module in six.iteritems(all_)
+  }
+
+
+def test():
+  """Returns dict of testing modules."""
+  return {name: module.test() for name, module in six.iteritems(all_)}
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  return {name: module.test_extra() for name, module in six.iteritems(all_)}
diff --git a/mathematics_dataset/mathematics_dataset/modules/numbers.py b/mathematics_dataset/mathematics_dataset/modules/numbers.py
new file mode 100644
index 0000000000000000000000000000000000000000..f261fdc481ab69dbc1a71b696d14facecec1dfaa
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/numbers.py
@@ -0,0 +1,516 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Number-related questions, e.g., "write seventy-two as a number"."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import math
+import random
+
+# Dependency imports
+import example
+from sample import number
+from util import composition
+from util import display
+import numpy as np
+import six
+from six.moves import range
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+_ENTROPY_EXTRAPOLATE = (12, 12)
+
+
+# Number of module compositions appearing in train/test, and extrapolation data.
+_NUM_MODULES_COMPOSED = [2, 4]
+
+
+def _make_modules(entropy, num_modules_composed):
+  """Returns modules given "difficulty" parameters."""
+  fns = {
+      'gcd': gcd,
+      'lcm': lcm,
+      'div_remainder': div_remainder,
+      'is_prime': is_prime,
+      'is_factor': is_factor,
+      'round_number': round_number,
+      'place_value': place_value,
+      'list_prime_factors': list_prime_factors,
+  }
+
+  # These modules don't have both pure and composed.
+  modules = {
+      'base_conversion': functools.partial(base_conversion, *entropy),
+  }
+
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  sample_args_composed = composition.PreSampleArgs(
+      num_modules_composed[0], num_modules_composed[1], *entropy)
+
+  for name, module in six.iteritems(fns):
+    modules[name] = functools.partial(module, None, sample_args_pure)
+    modules[name + '_composed'] = functools.partial(
+        module, None, sample_args_composed)
+
+  return modules
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(
+      entropy=entropy_fn(_ENTROPY_TRAIN),
+      num_modules_composed=_NUM_MODULES_COMPOSED)
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(
+      entropy=_ENTROPY_INTERPOLATE,
+      num_modules_composed=_NUM_MODULES_COMPOSED)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *_ENTROPY_EXTRAPOLATE)
+  return {
+      'round_number_big': functools.partial(
+          round_number, None, sample_args_pure),
+      'place_value_big': functools.partial(place_value, None, sample_args_pure),
+  }
+
+
+def place_value(value, sample_args, context=None):
+  """E.g., "Q: What is the tens digit of 31859? A: 5."""
+  del value  # unused for now
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  integer = number.integer(entropy, signed=False, min_abs=1)
+  (entity,) = context.sample(sample_args, [integer])
+
+  integer_as_string = str(integer)
+  num_digits = len(integer_as_string)
+
+  firsts = ['', 'dezena ', 'centena ']
+  seconds = [
+      'milhares', 'milhões', 'mil milhões', 'biliões', 'mil biliões',
+      'quinquiliões', 'sextilhões', 'septiliões', 'octiliões', 'nonalhões',
+      'decilhões',
+  ]
+  place_names = ['unidades', 'dezenas', 'centenas']
+  for second in seconds:
+    for first in firsts:
+      place_names.append(first + second)
+
+  place = random.randint(1, num_digits)  # 1 = units, 2 = tens, etc.
+  place_name = place_names[place - 1]
+  answer = sympy.Integer(integer_as_string[num_digits - place])
+
+  return example.Problem(
+      question=example.question(
+          context,
+          'Qual é o dígito das {place_name} de {integer}?',
+          place_name=place_name, integer=entity.expression_else_handle),
+      answer=answer)
+
+
+# TODO(b/124040078): add to composition system?
+def round_number(value, sample_args, context=None):
+  """Question for rounding integers and decimals."""
+  del value  # unused for now
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  # This is the power of 10 to round to. E.g., power == 0 corresponds to
+  # rounding to the nearest integer; power == -2 corresponds to rounding to two
+  # decimal places, and power == 3 corresponds to rounding to the nearest 1000.
+  power = random.randint(-7, 6)
+
+  answer_entropy = 1 + random.uniform(0, entropy / 2)
+  entropy = max(1, entropy - answer_entropy)
+  value_integer = number.integer(answer_entropy, signed=True)
+
+  remainder_divisor = 10 ** int(math.ceil(entropy))
+  remainder_range_lower = -remainder_divisor / 2
+  remainder_range_upper = remainder_divisor / 2
+
+  if value_integer <= 0:
+    remainder_range_lower += 1
+  if value_integer >= 0:
+    remainder_range_upper -= 1
+
+  remainder = random.randint(int(remainder_range_lower), int(remainder_range_upper))
+  input_ = value_integer + sympy.Rational(remainder, remainder_divisor)
+  scale = 10**power if power >= 0 else sympy.Rational(1, 10**(-power))
+  input_ = input_ * scale
+  value = value_integer * scale
+  if not number.is_integer(input_):
+    input_ = display.Decimal(input_)
+  if not number.is_integer(value):
+    value = display.Decimal(value)
+
+  (input_,) = context.sample(sample_args, [input_])
+
+  if power > 0:
+    # Rounding to a power of ten.
+    round_to = 10**power
+    if random.choice([False, True]):
+      # Write the rounding value as a word instead.
+      round_to = display.StringNumber(round_to,
+                                      join_number_words_with_hyphens=False)
+    description = '{round_to} mais próxima'.format(round_to=round_to)
+  elif power == 0 and random.choice([False, True]):
+    # Round to nearest integer.
+    description = 'inteiro mais próximo'
+  else:
+    # Round to decimal places.
+    description = 'casa decimal das {dps}'
+    if power != -1:
+      # Plural
+      description += 's'
+    dps = -power
+    if random.choice([False, True]):
+      dps = display.StringNumber(dps)
+    description = description.format(dps=dps)
+
+  template = random.choice([
+      'Arredonda {input} para as {description}.',
+      'Quanto é {input} arredondado para as {description}?',
+  ])
+
+  return example.Problem(
+      question=example.question(
+          context, template, input=input_, description=description),
+      answer=value)
+
+
+def _semi_prime(entropy):
+  """Generates a semi-prime with the given entropy."""
+  # Add on extra entropy to account for the sparsity of the primes; we don't
+  # actually use the integers sampled, but rather a random prime close to them;
+  # thus some entropy is lost, which we must account for
+  entropy += math.log10(max(1, entropy * math.log(10)))
+
+  # We intentionally uniformy sample the "entropy" (i.e., approx number digits)
+  # of the two factors.
+  entropy_1, entropy_2 = entropy * np.random.dirichlet([1, 1])
+
+  # Need >= 2 for randprime to always work (Betrand's postulate).
+  approx_1 = number.integer(entropy_1, signed=False, min_abs=2)
+  approx_2 = number.integer(entropy_2, signed=False, min_abs=2)
+
+  factor_1 = sympy.ntheory.generate.randprime(approx_1 / 2, approx_1 * 2)
+  factor_2 = sympy.ntheory.generate.randprime(approx_2 / 2, approx_2 * 2)
+
+  return factor_1 * factor_2
+
+
+def is_prime(value, sample_args, context=None):
+  """Questions asking about primality."""
+  del value  # unused for now
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  composite = _semi_prime(entropy)
+
+  if random.choice([False, True]):
+    # Use the composite
+    integer = composite
+    is_prime_ = False
+  else:
+    # Take the next prime after the composite, to ensure the same distribution
+    # as composites. Do "composite - 4" so we occasionally see "2" as a prime.
+    integer = sympy.ntheory.generate.nextprime(composite - 4)
+    is_prime_ = True
+
+  (integer_entity,) = context.sample(sample_args, [integer])
+
+  if random.choice([False, True]) and integer != 1:
+    answer = not is_prime_
+    attribute_name = random.choice(['compósito', 'um número compósito'])
+  else:
+    answer = is_prime_
+    attribute_name = random.choice(['primo', 'um número primo'])
+
+  return example.Problem(
+      question=example.question(
+          context, 'Será {integer} {attribute}?',
+          integer=integer_entity.expression_else_handle,
+          attribute=attribute_name),
+      answer=answer)
+
+
+def is_factor(value, sample_args, context=None):
+  """E.g., "Is 5 a factor of 48?"."""
+  del value  # unused
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  entropy_factor = 1 + random.uniform(0, entropy/3)
+  entropy = max(0, entropy - entropy_factor)
+  maybe_factor = number.integer(entropy_factor, False, min_abs=2)
+
+  integer = maybe_factor * number.integer(entropy, False, min_abs=1)
+  # Produce balanced classes.
+  if random.choice([False, True]):
+    # The following makes it not a factor.
+    integer += random.randint(1, maybe_factor - 1)
+
+  (entity,) = context.sample(sample_args, [integer])
+
+  templates = [
+      'Será {maybe_factor} um fator de {value}?',
+      'Será {value} um múltiplo de {maybe_factor}?',
+      'Será que {maybe_factor} divide {value}?',
+  ]
+  if maybe_factor == 2:
+    templates += [
+        '{value} é par?',
+    ]
+  template = random.choice(templates)
+
+  answer = integer % maybe_factor == 0
+  return example.Problem(
+      question=example.question(
+          context, template, maybe_factor=maybe_factor,
+          value=entity.expression_else_handle),
+      answer=answer)
+
+
+def list_prime_factors(value, sample_args, context=None):
+  """E.g., "What are the prime factors of 36?"."""
+  del value  # unused for now
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  entropy = max(1, entropy)
+
+  integer = number.integer(entropy, signed=False, min_abs=2)
+
+  (entity,) = context.sample(sample_args, [integer])
+  prime_factors = sorted(sympy.factorint(integer).keys())
+  template = random.choice([
+      'Quais são os fatores primos {integer}?',
+      'Enumera os fatores primos de {integer}.',
+  ])
+  return example.Problem(
+      question=example.question(
+          context, template, integer=entity.expression_else_handle),
+      answer=display.NumberList(prime_factors))
+
+
+def _pair_with_large_hidden_factor(entropy):
+  """Returns pair of numbers with possibly large common factor hidden."""
+  entropy_p, entropy_q, _ = entropy * np.random.dirichlet([1, 1, 1])
+  # Min entropy on p and q to minimize trivial solutions.
+  entropy_p = max(1, entropy_p)
+  entropy_q = max(1, entropy_q)
+  entropy_mult = max(0, entropy - entropy_p - entropy_q)
+
+  p = number.integer(entropy_p, False, min_abs=1)
+  q = number.integer(entropy_q, False, min_abs=1)
+  mult = number.integer(entropy_mult, False, min_abs=1)
+  p *= mult
+  q *= mult
+  return p, q
+
+
+def lcm(value, sample_args, context=None):
+  """Question for least common multiple of p and q."""
+  del value  # unused
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  p, q = _pair_with_large_hidden_factor(entropy)
+  answer = sympy.lcm(p, q)
+
+  if random.choice([False, True]):
+    p, q = context.sample(sample_args, [p, q])
+    # Ask the question directly.
+    adjective = 'menor'
+    template = random.choice([
+        'Calcula o {adjective} múltiplo comum de {p} e {q}.',
+        'Qual é o {adjective} múltiplo comum de {p} e {q}?',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, adjective=adjective, p=p.expression_else_handle,
+            q=q.expression_else_handle),
+        answer=answer)
+  else:
+    # Phrase the question as finding the common denominator of two fractions.
+    p = number.integer(2, signed=True, coprime_to=p) / p
+    q = number.integer(2, signed=True, coprime_to=q) / q
+    p, q = context.sample(sample_args, [p, q])
+
+    template = random.choice([
+        'Qual é o denominador comum de {p} e {q}?',
+        'Encontra o denominador comum de {p} e {q}.',
+        'Calcula o denominador comum de {p} e {q}.',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, p=p.expression_else_handle,
+            q=q.expression_else_handle),
+        answer=answer)
+
+
+def _random_coprime_pair(entropy):
+  """Returns a pair of random coprime integers."""
+  coprime_product = number.integer(entropy, False, min_abs=1)
+  factors = sympy.factorint(coprime_product)
+  def take():
+    prime = random.choice(list(factors.keys()))
+    power = factors[prime]
+    del factors[prime]
+    return prime ** power
+
+  if random.random() < 0.8 and len(factors) >= 2:
+    # Disallow trivial factoring where possible.
+    count_left = random.randint(1, len(factors) - 1)
+    count_right = len(factors) - count_left
+  else:
+    count_left = random.randint(0, len(factors))
+    count_right = len(factors) - count_left
+
+  left = sympy.prod([take() for _ in range(count_left)])
+  right = sympy.prod([take() for _ in range(count_right)])
+  assert left * right == coprime_product
+  return left, right
+
+
+# @composition.module(number.is_positive_integer)
+def gcd(value, sample_args, context=None):
+  """Question for greatest common divisor of p and q."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  if value is None:
+    value_entropy = 1 + random.uniform(0, entropy/3)
+    entropy = max(1, entropy - value_entropy)
+    value = number.integer(value_entropy, False, min_abs=1)
+
+  p_mult, q_mult = _random_coprime_pair(entropy)
+
+  p = value * p_mult
+  q = value * q_mult
+  assert sympy.gcd(p, q) == value
+
+  p, q = context.sample(sample_args, [p, q])
+
+  adjective = (random.choice(['maior ']) 
+               + random.choice(['divisor', 'fator']) + ' comum ')
+
+  if is_question:
+    template = random.choice([
+        'Calcula o {adjective} de {p} e {q}.',
+        'Qual é o {adjective} de {p} e {q}?',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, adjective=adjective, p=p, q=q),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        description='Seja {self}, o {adjective} de {p} e {q}.',
+        adjective=adjective, p=p, q=q)
+
+
+# @composition.module(number.is_positive_integer)
+def div_remainder(value, sample_args, context=None):
+  """E.g., "What is the remainder when 27 is divided by 5?"."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  if value is None:
+    entropy_value = 1 + random.uniform(0, entropy/3)
+    entropy = max(0, entropy - entropy_value)
+    value = number.integer(entropy_value, signed=False)
+
+  entropy_a, entropy_q = entropy * np.random.dirichlet([1, 1])
+  a = number.integer(entropy_a, signed=False, min_abs=1)
+  q = value + number.integer(entropy_q, signed=False, min_abs=1)
+
+  p = a * q + value
+  assert p % q == value
+  p, q = context.sample(sample_args, [p, q])
+
+  if is_question:
+    template = random.choice([
+        'Calcula o resto da divisão entre {p} e {q}.',
+        'Qual é o resto da divisão entre {p} e {q}?',
+    ])
+    return example.Problem(
+        question=example.question(
+            context, template, p=p.expression_else_handle,
+            q=q.expression_else_handle),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        description='Seja {self}, o resto da divisão entre {p} e {q}.',
+        p=p, q=q)
+
+
+def base_conversion(min_entropy, max_entropy):
+  """E.g., "What is 17 base 8 in base 10?"."""
+  context = composition.Context()
+
+  from_base = random.randint(2, 16)
+  while True:
+    to_base = random.randint(2, 16)
+    if to_base != from_base:
+      break
+
+  # Entropy used up in selecting bases.
+  entropy_used = math.log10(16 * 15)
+  entropy = random.uniform(
+      min_entropy - entropy_used, max_entropy - entropy_used)
+
+  value = number.integer(entropy, signed=True)
+  template = random.choice([
+      '{from_str} (em base {from_base}) para base {to_base}',
+      'Converte {from_str} (em base {from_base}) para base {to_base}.',
+      'Quanto é {from_str} (base {from_base}) em base {to_base}?',
+  ])
+  return example.Problem(
+      question=example.question(
+          context, template,
+          from_str=display.NumberInBase(value, from_base),
+          from_base=from_base,
+          to_base=to_base),
+      answer=display.NumberInBase(value, to_base))
diff --git a/mathematics_dataset/mathematics_dataset/modules/polynomials.py b/mathematics_dataset/mathematics_dataset/modules/polynomials.py
new file mode 100644
index 0000000000000000000000000000000000000000..8890e4267e9f340358f0e0d0cc6c8408ef4fb675
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/polynomials.py
@@ -0,0 +1,376 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Polynomial manipulation (adding, composing, finding coefficients, etc)."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import functools
+import math
+import random
+
+# Dependency imports
+import example
+from sample import number
+from sample import ops
+from sample import polynomials
+from util import composition
+import numpy as np
+from six.moves import range
+import sympy
+
+
+_ENTROPY_TRAIN = (3, 10)
+_ENTROPY_INTERPOLATE = (8, 8)
+
+
+def _make_modules(entropy):
+  """Returns modules given "difficulty" parameters."""
+  sample_args_pure = composition.PreSampleArgs(1, 1, *entropy)
+  sample_args_composed = composition.PreSampleArgs(2, 4, *entropy)
+  sample_args_mixed = composition.PreSampleArgs(1, 4, *entropy)
+
+  return {
+      'coefficient_named':
+          functools.partial(coefficient_named, None, sample_args_pure),
+      'evaluate':
+          functools.partial(evaluate, None, sample_args_pure),
+      'evaluate_composed':
+          functools.partial(evaluate, None, sample_args_composed),
+      # TODO(b/124038948): consider doing pure sample args for 'add'?
+      'add':
+          functools.partial(add, None, sample_args_mixed),
+      'expand':
+          functools.partial(expand, None, sample_args_pure),
+      'collect':
+          functools.partial(collect, None, sample_args_pure),
+      'compose':
+          functools.partial(compose, None, sample_args_mixed),
+
+      # Rearranging powers:
+      'simplify_power':
+          functools.partial(simplify_power, None, sample_args_pure),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  return _make_modules(entropy_fn(_ENTROPY_TRAIN))
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(_ENTROPY_INTERPOLATE)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  return {
+  }
+
+
+def coefficient_named(value, sample_args, context=None):
+  """E.g., "Express x^2 + 2x in the form h * x^2 + k * x + t and give h."."""
+  del value  # not used
+  if context is None:
+    context = composition.Context()
+  variable = sympy.Symbol(context.pop())
+
+  entropy, sample_args = sample_args.peel()
+  degree = random.randint(1, 4)
+  if random.choice([False, True]):
+    coefficients = polynomials.sample_coefficients(
+        degree, entropy/2, min_non_zero=random.randint(degree - 1, degree))
+    expanded = polynomials.expand_coefficients(coefficients, entropy/2)
+    expression = polynomials.coefficients_to_polynomial(expanded, variable)
+  else:
+    expression = polynomials.sample_with_brackets(variable, degree, entropy)
+    coefficients = list(reversed(sympy.Poly(expression).all_coeffs()))
+
+  named_coeffs = [sympy.Symbol(context.pop()) for _ in range(degree + 1)]
+  canonical = polynomials.coefficients_to_polynomial(named_coeffs, variable)
+
+  if random.random() < 0.2:  # only small probability of non-zero power
+    power = random.randint(0, degree)
+  else:
+    non_zero_powers = [i for i in range(degree + 1) if coefficients[i] != 0]
+    power = random.choice(non_zero_powers)
+
+  value = coefficients[power]
+  named_coeff = named_coeffs[power]
+
+  template = random.choice([
+      'Expressa {expression} como {canonical} e dá {target}.',
+      'Rearranja {expression} como {canonical} e dá {target}.',
+      'Expressa {expression} na forma {canonical} e dá {target}.',
+      'Rearranja {expression} na forma {canonical} e dá {target}.',
+  ])
+  return example.Problem(
+      question=example.question(
+          context, template, expression=expression, canonical=canonical,
+          target=named_coeff),
+      answer=value)
+
+
+_TEMPLATES = [
+    'O que é {composed}?',
+    'Calcula {composed}.',
+    'Obtém {composed}.',
+    'Determina {composed}.',
+]
+
+
+@composition.module(number.is_integer)
+def evaluate(value, sample_args, context=None):
+  """Entity for evaluating an integer-valued polynomial at a given point."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  if value is None:
+    entropy_value = random.uniform(1, 1 + entropy/3)
+    entropy = max(0, entropy - entropy_value)
+    value = number.integer(entropy_value, signed=True)
+
+  entropy_input = random.uniform(1, 1 + entropy/3)
+  entropy = max(0, entropy - entropy_input)
+  input_ = number.integer(entropy_input, signed=True)
+
+  degree = random.randint(1, 3)
+
+  entropies = entropy * np.random.dirichlet(list(range(1, degree + 1)))
+  # Calculate coefficients in reverse order.
+  target = value
+  coeffs_reversed = []
+  for i, coeff_entropy in enumerate(entropies):
+    power = degree - i
+    coeff = number.integer(coeff_entropy, signed=True)
+    if input_ != 0:
+      coeff += int(round(target / input_ ** power))
+    if coeff == 0 and i == 0:
+      # Don't allow zero in leading coefficient.
+      coeff += random.choice([-1, 1])
+    coeffs_reversed.append(coeff)
+    target -= coeff * (input_ ** power)
+  coeffs_reversed.append(target)
+
+  coefficients = list(reversed(coeffs_reversed))
+
+  (polynomial_entity, input_) = context.sample(
+      sample_args, [composition.Polynomial(coefficients), input_])
+  composed = polynomial_entity.handle.apply(input_.handle)
+
+  if is_question:
+    template = random.choice(_TEMPLATES)
+    return example.Problem(
+        question=example.question(context, template, composed=composed),
+        answer=value)
+  else:
+    return composition.Entity(
+        context=context,
+        value=value,
+        expression=composed,
+        description='Seja {self}, {composed}.',
+        composed=composed)
+
+
+# TODO(b/124039290): merge with compose? both add and compose do similar things.
+@composition.module(composition.is_integer_polynomial)
+def add(value, sample_args, context=None):
+  """E.g., "Let f(x)=2x+1, g(x)=3x+2. What is 5*f(x) - 7*g(x)?"."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  if value is None:
+    max_degree = 3
+    degree = random.randint(1, max_degree)
+    entropy -= math.log10(max_degree)
+    entropy_value = entropy / 2
+    entropy -= entropy_value
+    value = polynomials.sample_coefficients(
+        degree, entropy=entropy_value, min_non_zero=random.randint(1, 3))
+    value = composition.Polynomial(value)
+
+  c1, c2, coeffs1, coeffs2 = polynomials.coefficients_linear_split(
+      value.coefficients, entropy)
+  coeffs1 = polynomials.trim(coeffs1)
+  coeffs2 = polynomials.trim(coeffs2)
+
+  c1, c2, fn1, fn2 = context.sample(
+      sample_args,
+      [c1, c2, composition.Polynomial(coeffs1), composition.Polynomial(coeffs2)]
+  )
+
+  var = sympy.var(context.pop())
+
+  expression = (
+      c1.handle * fn1.handle.apply(var) + c2.handle * fn2.handle.apply(var))
+
+  if is_question:
+    answer = polynomials.coefficients_to_polynomial(value.coefficients, var)
+    answer = answer.sympy()
+    template = random.choice(_TEMPLATES)
+    return example.Problem(
+        question=example.question(context, template, composed=expression),
+        answer=answer)
+  else:
+    intermediate_symbol = context.pop()
+    intermediate = sympy.Function(intermediate_symbol)(var)
+    return composition.Entity(
+        context=context,
+        value=value,
+        description='Seja {intermediate} = {composed}.',
+        handle=composition.FunctionHandle(intermediate_symbol),
+        intermediate=intermediate,
+        composed=expression)
+
+
+def expand(value, sample_args, context=None):
+  """E.g., "Expand (x**2 + 1)**2."."""
+  del value  # not used
+  if context is None:
+    context = composition.Context()
+  variable = sympy.Symbol(context.pop())
+  entropy, sample_args = sample_args.peel()
+
+  min_order = 1
+  max_order = 5
+  order = random.randint(min_order, max_order)
+  entropy -= math.log10(max_order - min_order + 1)
+  expression_ = polynomials.sample_with_brackets(variable, order, entropy)
+  expanded = sympy.expand(expression_)
+  template = random.choice([
+      'Expande {expression}.'
+  ])
+  return example.Problem(
+      question=example.question(context, template, expression=expression_),
+      answer=expanded)
+
+
+@composition.module(composition.is_polynomial)
+def collect(value, sample_args, context=None):
+  """Collect terms in an unsimplified polynomial."""
+  is_question = context is None
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  if value is None:
+    entropy_value, entropy = entropy * np.random.dirichlet([2, 3])
+    degrees = [random.randint(1, 3)]
+    value = composition.Polynomial(
+        polynomials.sample_coefficients(degrees, entropy_value))
+
+  assert isinstance(value, composition.Polynomial)
+  coefficients = value.coefficients
+
+  all_coefficients_are_integer = True
+  for coeff in coefficients.flat:
+    if not number.is_integer(coeff):
+      all_coefficients_are_integer = False
+      break
+
+  if all_coefficients_are_integer:
+    coefficients = polynomials.expand_coefficients(coefficients, entropy)
+  else:
+    # put back the unused entropy
+    sample_args = composition.SampleArgs(
+        sample_args.num_modules, sample_args.entropy + entropy)
+
+  num_variables = coefficients.ndim
+  variables = [sympy.Symbol(context.pop()) for _ in range(num_variables)]
+  unsimplified = polynomials.coefficients_to_polynomial(coefficients, variables)
+  simplified = unsimplified.sympy().expand()
+
+  # Bit of a hack: handle the very rare case where no number constants appearing
+  if not ops.number_constants(unsimplified):
+    unsimplified = ops.Add(unsimplified, ops.Constant(0))
+  context.sample_by_replacing_constants(sample_args, unsimplified)
+
+  if is_question:
+    template = 'Coleta os termos em {unsimplified}.'
+    return example.Problem(
+        question=example.question(context, template, unsimplified=unsimplified),
+        answer=simplified)
+  else:
+    function_symbol = context.pop()
+    function = sympy.Function(function_symbol)(*variables)
+    return composition.Entity(
+        context=context,
+        value=value,
+        handle=composition.FunctionHandle(function_symbol),
+        expression=unsimplified,
+        polynomial_variables=variables,
+        description='Seja {function} = {unsimplified}.',
+        function=function,
+        unsimplified=unsimplified)
+
+
+def compose(value, sample_args, context=None):
+  """E.g., "Let f(x)=2x+1, let g(x)=3x+10. What is f(g(x))?"."""
+  del value  # unused
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+  entropy_f, entropy_g = entropy * np.random.dirichlet([1, 1])
+
+  coeffs_f = polynomials.sample_coefficients([random.randint(1, 2)], entropy_f)
+  coeffs_g = polynomials.sample_coefficients([random.randint(1, 2)], entropy_g)
+
+  entity_f, entity_g = context.sample(
+      sample_args,
+      [composition.Polynomial(coeffs_f), composition.Polynomial(coeffs_g)])
+
+  variable = sympy.var(context.pop())
+
+  poly_f = polynomials.coefficients_to_polynomial(coeffs_f, variable)
+  poly_g = polynomials.coefficients_to_polynomial(coeffs_g, variable)
+
+  poly_f_g = poly_f.sympy().subs(variable, poly_g.sympy()).expand()
+
+  expression = composition.FunctionHandle(entity_f, entity_g).apply(variable)
+
+  template = random.choice(_TEMPLATES)
+  return example.Problem(
+      question=example.question(context, template, composed=expression),
+      answer=poly_f_g)
+
+
+def simplify_power(value, sample_args, context=None):
+  """E.g., "Simplify ((x**2)**3/x**4)**2/x**3."."""
+  del value  # unused
+  if context is None:
+    context = composition.Context()
+
+  entropy, sample_args = sample_args.peel()
+
+  variable = sympy.symbols(context.pop(), positive=True)
+  unsimplified = polynomials.sample_messy_power(variable, entropy)
+  answer = unsimplified.sympy()
+
+  template = random.choice([
+      'Simplifica {unsimplified} assumindo que {variable} é positiva.',
+  ])
+  return example.Problem(
+      example.question(
+          context, template, unsimplified=unsimplified, variable=variable),
+      answer)
diff --git a/mathematics_dataset/mathematics_dataset/modules/probability.py b/mathematics_dataset/mathematics_dataset/modules/probability.py
new file mode 100644
index 0000000000000000000000000000000000000000..5982ddeb0c50c4c0aab90700e8bba4c5174460c8
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/probability.py
@@ -0,0 +1,262 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Probability questions (sampling, independence, expectations, ...)."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+import functools
+import random
+import string
+
+# Dependency imports
+import example
+from modules import train_test_split
+from util import combinatorics
+from util import composition
+from util import display
+from util import probability
+import numpy as np
+from six.moves import range
+from six.moves import zip
+
+
+_LETTERS = string.ascii_lowercase
+
+_MAX_FRAC_TRIVIAL_PROB = 0.1
+
+# Maximum number of colours and objects in a bag.
+_MAX_DISTINCT_LETTERS = 6
+_MAX_TOTAL_LETTERS = 20
+_MAX_LETTER_REPEAT = 10
+
+_SWR_SAMPLE_COUNT = [2, 4]
+_SWR_SAMPLE_COUNT_EXTRAPOLATE = [5, 5]
+
+_GERUNDS = {
+    'escolher': 'escolher',
+}
+
+
+def _make_modules(is_train):
+  """Returns modules, with split based on the boolean `is_train`."""
+  return {
+      'swr_p_sequence': functools.partial(
+          swr_prob_sequence, is_train=is_train, sample_range=_SWR_SAMPLE_COUNT),
+      'swr_p_level_set': functools.partial(
+          swr_prob_level_set, is_train=is_train,
+          sample_range=_SWR_SAMPLE_COUNT),
+  }
+
+
+def train(entropy_fn):
+  """Returns dict of training modules."""
+  del entropy_fn  # unused
+  return _make_modules(is_train=True)
+
+
+def test():
+  """Returns dict of testing modules."""
+  return _make_modules(is_train=False)
+
+
+def test_extra():
+  """Returns dict of extrapolation testing modules."""
+  return {
+      'swr_p_sequence_more_samples': functools.partial(
+          swr_prob_sequence, is_train=None,
+          sample_range=_SWR_SAMPLE_COUNT_EXTRAPOLATE),
+      'swr_p_level_set_more_samples': functools.partial(
+          swr_prob_level_set, is_train=None,
+          sample_range=_SWR_SAMPLE_COUNT_EXTRAPOLATE),
+  }
+
+
+def _sequence_event(values, length, verb):
+  """Returns sequence (finite product) event.
+
+  Args:
+    values: List of values to sample from.
+    length: Length of the sequence to generate.
+    verb: Verb in infinitive form.
+
+  Returns:
+    Instance of `probability.FiniteProductEvent`, together with a text
+    description.
+  """
+  del verb  # unused
+  samples = [random.choice(values) for _ in range(length)]
+  events = [probability.DiscreteEvent([sample]) for sample in samples]
+  event = probability.FiniteProductEvent(events)
+  sequence = ''.join(str(sample) for sample in samples)
+  event_description = 'sequência {sequence}'.format(sequence=sequence)
+  return event, event_description
+
+
+def _word_series(words, conjunction='e'):
+  """Combines the words using commas and the final conjunction."""
+  len_words = len(words)
+  if len_words == 0:
+    return ''
+  if len_words == 1:
+    return words[0]
+  return '{} {} {}'.format(', '.join(words[:-1]), conjunction, words[-1])
+
+
+def _level_set_event(values, length, verb):
+  """Generates `LevelSetEvent`; see _generate_sequence_event."""
+  counts = combinatorics.uniform_non_negative_integers_with_sum(
+      len(values), length)
+  counts_dict = dict(list(zip(values, counts)))
+  event = probability.CountLevelSetEvent(counts_dict)
+
+  shuffled_values = list(values)
+  random.shuffle(shuffled_values)
+
+  counts_and_values = [
+      '{} {}'.format(counts_dict[value], value)
+      for value in shuffled_values
+      if counts_dict[value] > 0
+  ]
+  counts_and_values = _word_series(counts_and_values)
+  template = random.choice([
+      '{verbing} {counts_and_values}',
+  ])
+  verbing = _GERUNDS[verb]
+  event_description = template.format(
+      counts_and_values=counts_and_values, verbing=verbing)
+  return event, event_description
+
+
+LetterBag = collections.namedtuple(
+    'LetterBag',
+    ('weights', 'random_variable', 'letters_distinct', 'bag_contents'))
+
+
+def _sample_letter_bag(is_train, min_total):
+  """Samples a "container of letters" and returns info on it."""
+  while True:
+    num_distinct_letters = random.randint(1, _MAX_DISTINCT_LETTERS)
+    num_letters_total = random.randint(
+        max(num_distinct_letters, min_total),
+        min(_MAX_TOTAL_LETTERS, num_distinct_letters * _MAX_LETTER_REPEAT))
+    letter_counts = combinatorics.uniform_positive_integers_with_sum(
+        num_distinct_letters, num_letters_total)
+
+    # Test/train split.
+    if (is_train is None
+        or train_test_split.is_train(sorted(letter_counts)) == is_train):
+      break
+
+  letters_distinct = random.sample(_LETTERS, num_distinct_letters)
+  weights = {i: 1 for i in range(num_letters_total)}
+
+  letters_with_repetition = []
+  for letter, count in zip(letters_distinct, letter_counts):
+    letters_with_repetition += [letter] * count
+  random.shuffle(letters_with_repetition)
+
+  random_variable = probability.DiscreteRandomVariable(
+      {i: letter for i, letter in enumerate(letters_with_repetition)})
+
+  if random.choice([False, True]):
+    bag_contents = ''.join(letters_with_repetition)
+  else:
+    letters_and_counts = [
+        '{}: {}'.format(letter, count)
+        for letter, count in zip(letters_distinct, letter_counts)]
+    bag_contents = '{' + ', '.join(letters_and_counts) + '}'
+
+  return LetterBag(
+      weights=weights,
+      random_variable=random_variable,
+      letters_distinct=letters_distinct,
+      bag_contents=bag_contents)
+
+
+def _swr_space(is_train, sample_range):
+  """Returns probability space for sampling without replacement."""
+  num_sampled = random.randint(*sample_range)
+  sample = _sample_letter_bag(is_train=is_train, min_total=num_sampled)
+
+  space = probability.SampleWithoutReplacementSpace(sample.weights, num_sampled)
+
+  random_variable = probability.FiniteProductRandomVariable(
+      [sample.random_variable] * num_sampled)
+
+  random_variable.description = (
+      str(display.StringNumber(num_sampled))
+      + ' caractéres são escolhidos sem reposição de '
+      + sample.bag_contents)
+
+  return sample.letters_distinct, space, random_variable
+
+
+def _sample_without_replacement_probability_question(
+    is_train, event_fn, sample_range):
+  """Question for prob of some event when sampling without replacement."""
+  def too_big(event_in_space):
+    if isinstance(event_in_space, probability.SequenceEvent):
+      size = len(event_in_space.all_sequences())
+    else:
+      assert isinstance(event_in_space, probability.FiniteProductEvent)
+      size = np.prod([len(event.values) for event in event_in_space.events])
+    return size > int(2e5)
+
+  allow_trivial_prob = random.random() < _MAX_FRAC_TRIVIAL_PROB
+
+  while True:
+    distinct_letters, space, random_variable = _swr_space(
+        is_train, sample_range)
+
+    event, event_description = event_fn(
+        values=distinct_letters, length=space.n_samples, verb='escolher')
+    event_in_space = random_variable.inverse(event)
+    if too_big(event_in_space):
+      continue
+    answer = space.probability(event_in_space)
+    if answer not in [0, 1] or allow_trivial_prob:
+      break
+
+  context = composition.Context()
+
+  template = random.choice([
+      '{random_variable_capitalize}. Qual é a probabilidade de(a) {event}?',
+      '{random_variable_capitalize}. Dá a probabilidade de(a) {event}.',
+      'Qual é a probabilidade de(a) {event} quando {random_variable}?',
+      'Calcula da probabilidade de(a) {event} quando {random_variable}.',
+  ])
+  question = example.question(
+      context,
+      template,
+      random_variable=random_variable.description,
+      random_variable_capitalize=(
+          str(random_variable.description).capitalize()),
+      event=event_description)
+  return example.Problem(question, answer)
+
+
+def swr_prob_sequence(is_train, sample_range):
+  """Probability of given sequence when sampling without replacement."""
+  return _sample_without_replacement_probability_question(
+      is_train=is_train, event_fn=_sequence_event, sample_range=sample_range)
+
+
+def swr_prob_level_set(is_train, sample_range):
+  """Probability of given level set when sampling without replacement."""
+  return _sample_without_replacement_probability_question(
+      is_train=is_train, event_fn=_level_set_event, sample_range=sample_range)
diff --git a/mathematics_dataset/mathematics_dataset/modules/train_test_split.py b/mathematics_dataset/mathematics_dataset/modules/train_test_split.py
new file mode 100644
index 0000000000000000000000000000000000000000..e85d3d991a480d0e8ef6cddd31bd0ed49aff9d3a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/modules/train_test_split.py
@@ -0,0 +1,27 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Utility for train/test split based on hash value."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import hashlib
+
+
+def is_train(value):
+  """Returns whether `value` should be used in a training question."""
+  value_as_string = str(value).encode('utf-8')
+  return int(hashlib.md5(value_as_string).hexdigest(), 16) % 2 == 0
diff --git a/mathematics_dataset/mathematics_dataset/requires.txt b/mathematics_dataset/mathematics_dataset/requires.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ab0152c65b05ca59fe6c13c1e2ce42cc247ddce1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/requires.txt
@@ -0,0 +1,4 @@
+absl-py>=0.1.0
+numpy>=1.10
+six
+sympy>=1.2
diff --git a/mathematics_dataset/mathematics_dataset/sample/__init__.py b/mathematics_dataset/mathematics_dataset/sample/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..d780829a1f5101333577a1ca9adcb960529213f1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/__init__.py
@@ -0,0 +1,13 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/mathematics_dataset/mathematics_dataset/sample/arithmetic.py b/mathematics_dataset/mathematics_dataset/sample/arithmetic.py
new file mode 100644
index 0000000000000000000000000000000000000000..d439387c7e986c5d8278d0bc19a269e243ac2cb6
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/arithmetic.py
@@ -0,0 +1,260 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Sample arithmetic expressions with a given value."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+import math
+import random
+
+# Dependency imports
+from sample import number
+from sample import ops
+from util import combinatorics
+import numpy as np
+import six
+from six.moves import zip
+import sympy
+
+
+class _SampleArgs(collections.namedtuple('SampleArgs', ('count', 'entropy'))):
+  """For sampling mathematical expressions."""
+
+  def peel(self, frac=1):
+    """Peels one (or `frac`) of an op's entropy."""
+    entropy = frac * self.entropy / self.count
+    new_sample_args = _SampleArgs(self.count, self.entropy - entropy)
+    return entropy, new_sample_args
+
+  def split(self, args):
+    """Splits the entropy and op counts up."""
+    non_integer_count = sum(not arg.is_Integer for arg in args)
+    assert non_integer_count <= self.count - 1
+    count_split = combinatorics.uniform_non_negative_integers_with_sum(
+        len(args), (self.count - 1) - non_integer_count)
+    for i, arg in enumerate(args):
+      if not arg.is_Integer:
+        count_split[i] += 1
+    if all(count == 0 for count in count_split):
+      assert self.entropy == 0
+      entropies = np.zeros(len(count_split))
+    else:
+      entropies = (
+          np.random.dirichlet(np.maximum(1e-9, count_split)) * self.entropy)
+    return [_SampleArgs(op_count, entropy)
+            for op_count, entropy in zip(count_split, entropies)]
+
+
+def _add_sub_filter(value, sample_args):
+  return sample_args.count >= 2 or value.is_Integer
+
+
+def _add_op(value, sample_args, rationals_allowed):
+  """Returns sampled args for `ops.Add`."""
+  entropy, sample_args = sample_args.peel()
+  if rationals_allowed and sample_args.count >= 3:
+    x = number.integer_or_rational(entropy, True)
+  else:
+    x = number.integer(entropy, True)
+  if random.choice([False, True]):
+    op_args = [x, value - x]
+  else:
+    op_args = [value - x, x]
+  return ops.Add, op_args, sample_args
+
+
+def _sub_op(value, sample_args, rationals_allowed):
+  """Returns sampled args for `ops.Sub`."""
+  entropy, sample_args = sample_args.peel()
+  if rationals_allowed and sample_args.count >= 3:
+    x = number.integer_or_rational(entropy, True)
+  else:
+    x = number.integer(entropy, True)
+  if random.choice([False, True]):
+    op_args = [x, x - value]
+  else:
+    op_args = [value + x, x]
+  return ops.Sub, op_args, sample_args
+
+
+def _entropy_of_factor_split(integer):
+  """Returns entropy (log base 10) of decomposing: integer = a * b."""
+  assert integer.is_Integer
+  if integer == 0:
+    return 0
+  # Gives dict of form {factor: multiplicity}
+  factors = sympy.factorint(integer)
+  return sum(math.log10(mult + 1) for mult in six.itervalues(factors))
+
+
+def _split_factors(integer):
+  """Randomly factors integer into product of two integers."""
+  assert integer.is_Integer
+  if integer == 0:
+    return [1, 0]
+  # Gives dict of form {factor: multiplicity}
+  factors = sympy.factorint(integer)
+  left = sympy.Integer(1)
+  right = sympy.Integer(1)
+  for factor, mult in six.iteritems(factors):
+    left_mult = random.randint(0, mult)
+    right_mult = mult - left_mult
+    left *= factor ** left_mult
+    right *= factor ** right_mult
+  return left, right
+
+
+def _mul_filter(value, sample_args):
+  if sample_args.count >= 2:
+    return True
+  if not value.is_Integer:
+    return False
+  return sample_args.entropy <= _entropy_of_factor_split(value)
+
+
+def _mul_op(value, sample_args, rationals_allowed):
+  """Returns sampled args for `ops.Mul`."""
+  if sample_args.count >= 3:
+    _, op_args, sample_args = _div_op(value, sample_args, rationals_allowed)
+    op_args = [op_args[0], sympy.Integer(1) / op_args[1]]
+  elif sample_args.count == 1:
+    entropy, sample_args = sample_args.peel()
+    assert _entropy_of_factor_split(value) >= entropy
+    op_args = _split_factors(value)
+  else:
+    assert sample_args.count == 2
+    entropy, sample_args = sample_args.peel()
+    numer = sympy.numer(value)
+    denom = sympy.denom(value)
+    p1, p2 = _split_factors(numer)
+    entropy -= _entropy_of_factor_split(numer)
+    mult = number.integer(entropy, signed=True, min_abs=1, coprime_to=p1)
+    op_args = [p1 / (mult * denom), p2 * mult]
+
+  if random.choice([False, True]):
+    op_args = list(reversed(op_args))
+
+  return ops.Mul, op_args, sample_args
+
+
+def _div_filter(value, sample_args):
+  del value  # unused
+  del sample_args  # unused
+  return True
+
+
+def _div_op(value, sample_args, rationals_allowed):
+  """Returns sampled args for `ops.Div`."""
+  assert rationals_allowed  # should be True if this function gets invoked
+  entropy, sample_args = sample_args.peel()
+
+  numer = sympy.numer(value)
+  denom = sympy.denom(value)
+
+  if sample_args.count == 1:
+    mult = number.integer(entropy, signed=True, min_abs=1)
+    op_args = [numer * mult, denom * mult]
+  elif sample_args.count == 2:
+    if numer == 0 or random.choice([False, True]):
+      x = number.integer(entropy, signed=True, min_abs=1, coprime_to=denom)
+      op_args = [sympy.Rational(x * numer, denom), x]
+    else:
+      x = number.integer(entropy, signed=True, min_abs=1, coprime_to=numer)
+      op_args = [x, sympy.Rational(x * denom, numer)]
+  else:
+    assert sample_args.count >= 3
+    p2, p1 = _split_factors(numer)
+    q1, q2 = _split_factors(denom)
+    entropy -= _entropy_of_factor_split(numer) + _entropy_of_factor_split(denom)
+    entropy_r = random.uniform(0, entropy)
+    entropy_s = entropy - entropy_r
+    r = number.integer(entropy_r, signed=True, min_abs=1, coprime_to=q1*p2)
+    s = number.integer(entropy_s, signed=False, min_abs=1, coprime_to=p1*q2)
+    op_args = [sympy.Rational(r*p1, s*q1), sympy.Rational(r*q2, s*p2)]
+
+  return ops.Div, op_args, sample_args
+
+
+def _arithmetic(value, sample_args, add_sub, mul_div):
+  """Internal arithmetic thingy...."""
+  assert sample_args.count >= 0
+  if sample_args.count == 0:
+    assert sample_args.entropy == 0
+    return ops.Constant(value)
+
+  allowed = []
+  if add_sub and _add_sub_filter(value, sample_args):
+    allowed.append(_add_op)
+    allowed.append(_sub_op)
+  if mul_div and _mul_filter(value, sample_args):
+    allowed.append(_mul_op)
+  if mul_div and _div_filter(value, sample_args):
+    allowed.append(_div_op)
+  if not allowed:
+    raise ValueError(
+        'No valid ops found, add_sub={} mul_div={} value={} sample_args={}'
+        .format(add_sub, mul_div, value, sample_args))
+  choice = random.choice(allowed)
+
+  op, args, sample_args = choice(value, sample_args, rationals_allowed=mul_div)
+  sample_args = sample_args.split(args)
+  child_expressions = [_arithmetic(arg, child_sample_arg, add_sub, mul_div)
+                       for arg, child_sample_arg in zip(args, sample_args)]
+
+  return op(*child_expressions)
+
+
+def length_range_for_entropy(entropy):
+  """Returns length range to sample from for given entropy."""
+  min_length = 3
+  max_length = min_length + int(entropy / 2)
+  return min_length, max_length
+
+
+def arithmetic(value, entropy, length=None, add_sub=True, mul_div=True):
+  """Generates an arithmetic expression with a given value.
+
+  Args:
+    value: Target value (integer or rational).
+    entropy: Amount of randomness to use in generating expression.
+    length: Number of ops to use. If `None` then suitable length will be picked
+        based on entropy by sampling within the range
+        `length_range_for_entropy`.
+    add_sub: Whether to include addition and subtraction operations.
+    mul_div: Whether to include multiplication and division operations.
+
+  Returns:
+    Instance of `ops.Op` containing expression.
+  """
+  assert isinstance(entropy, float)
+  if length is None:
+    min_length, max_length = length_range_for_entropy(entropy)
+    length = random.randint(min_length, max_length)
+    # Some entropy used up in sampling the length.
+    entropy -= math.log10(max_length - min_length + 1)
+  else:
+    assert isinstance(length, int)
+
+  # Entropy adjustment, because different binary trees (from sampling ops) can
+  # lead to the same expression. This is the correct value when we use just
+  # addition as the op, and is otherwise an an upper bound.
+  entropy += combinatorics.log_number_binary_trees(length) / math.log(10)
+
+  value = sympy.sympify(value)
+  sample_args = _SampleArgs(length, entropy)
+  return _arithmetic(value, sample_args, add_sub, mul_div)
diff --git a/mathematics_dataset/mathematics_dataset/sample/arithmetic_test.py b/mathematics_dataset/mathematics_dataset/sample/arithmetic_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..e19441f8989655223677bc7f8ccdbf73af85b2fd
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/arithmetic_test.py
@@ -0,0 +1,55 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.sample.arithmetic."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from absl.testing import absltest
+from absl.testing import parameterized
+from sample import arithmetic
+from sample import number
+from sample import ops
+from six.moves import range
+import sympy
+
+
+class ArithmeticTest(parameterized.TestCase):
+
+  def testArithmetic(self):
+    for _ in range(1000):
+      target = number.integer_or_rational(4, signed=True)
+      entropy = 8.0
+      expression = arithmetic.arithmetic(target, entropy)
+      self.assertEqual(sympy.sympify(expression), target)
+
+  def testArithmeticLength(self):
+    """Tests that the generated arithmetic expressions have given length."""
+    for _ in range(1000):
+      target = number.integer_or_rational(4, signed=True)
+      entropy = 8.0
+      length = random.randint(2, 10)
+      expression = arithmetic.arithmetic(target, entropy, length)
+      # Note: actual length is #ops = #numbers - 1.
+      actual_length = len(ops.number_constants(expression)) - 1
+      self.assertEqual(actual_length, length)
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/sample/linear_system.py b/mathematics_dataset/mathematics_dataset/sample/linear_system.py
new file mode 100644
index 0000000000000000000000000000000000000000..6e63307393975f9fc2e56a014d51035a8f61b028
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/linear_system.py
@@ -0,0 +1,144 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Generate linear systems with given set of solutions."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from sample import number
+from sample import ops
+from sample import polynomials
+import numpy as np
+from six.moves import range
+import sympy
+
+
+def _make_equals_zero_split(monomials):
+  """Returns an `ops.Eq` containing sum of monomials split on left and right."""
+  left = []
+  right = []
+  for monomial in monomials:
+    if random.choice([False, True]):
+      left.append(monomial)
+    else:
+      right.append(ops.Neg(monomial))
+  if not left:
+    left = [0]
+  if not right:
+    right = [0]
+  left = ops.Add(*left)
+  right = ops.Add(*right)
+  return ops.Eq(left, right)
+
+
+def _is_trivial_in(matrix, variable):
+  """Returns true if matrix_ij == 0 for some i and all j != variable."""
+  matrix = np.asarray(matrix)
+  assert matrix.ndim == 2 and matrix.shape[0] == matrix.shape[1]
+  size = matrix.shape[0]
+  if size == 1:
+    return False
+  for i in range(size):
+    all_zero = True
+    for j in range(size):
+      if j != variable and matrix[i, j] != 0:
+        all_zero = False
+        break
+    if all_zero:
+      return True
+  return False
+
+
+def _invertible_matrix(degree, entropy, non_trivial_in):
+  """Generates random invertible matrix."""
+  matrix_entropies = entropy * np.random.dirichlet(np.ones(degree * degree))
+  matrix_entropies = np.reshape(matrix_entropies, [degree, degree])
+  matrix_entropies = np.maximum(1, matrix_entropies)
+
+  while True:
+    def gen(i, j):
+      return number.integer(matrix_entropies[i, j], True)
+
+    matrix = [[gen(i, j) for i in range(degree)] for j in range(degree)]  # pylint: disable=g-complex-comprehension
+    if non_trivial_in is not None and _is_trivial_in(matrix, non_trivial_in):
+      continue
+    if sympy.det(sympy.Matrix(matrix)) != 0:
+      break
+
+  matrix = np.asarray(matrix).astype(int)
+  return matrix
+
+
+def linear_system(variables, solutions, entropy, non_trivial_in=None,
+                  length=None):
+  """Returns a linear system (set of equalities) with the given solutions.
+
+  Args:
+    variables: List of variables.
+    solutions: List of solutions, of the same length as `variables`.
+    entropy: Float >= 0; the entropy used.
+    non_trivial_in: Optional integer corresponding to a variable for which the
+        solution shouldn't be "trivial". E.g., "solve a + b = 3, a = -2 for a"
+        is disallowed if `variables[non_trivial_in] == 'a'`.
+    length: Total number of terms appearing; if `None` then selected wisely.
+
+  Returns:
+    List of `ops.Eq`.
+  """
+  degree = len(variables)
+  assert degree == len(solutions)
+
+  frac_entropy_matrix = random.uniform(1/3, 2/3)
+  matrix = _invertible_matrix(
+      degree, entropy * frac_entropy_matrix, non_trivial_in)
+  solutions = np.asarray(solutions)
+  constant = np.matmul(matrix, solutions.astype(int))
+  flattened = np.concatenate([np.reshape(matrix, [degree * degree]), constant])
+  is_zero = flattened == 0
+
+  if length is None:
+    min_length = np.count_nonzero(flattened) + 1
+    max_length = max(min_length, 1 + int(degree * (1 + entropy / 2)))
+    length = random.randint(min_length, max_length)
+
+  counts = polynomials.expanded_coefficient_counts(
+      length=length, is_zero=is_zero)
+
+  entropies = (1 - frac_entropy_matrix) * entropy * np.random.dirichlet(
+      np.maximum(1e-9, counts - 1))
+
+  terms = []
+  for i in range(len(flattened)):
+    coeffs = polynomials.integers_with_sum(
+        value=flattened[i], count=counts[i], entropy=entropies[i])
+    terms.append(coeffs)
+
+  matrix = terms[:degree*degree]
+  constant = terms[-degree:]
+  equations = []
+  for row_index in range(degree):
+    monomials = []
+    for col_index in range(degree):
+      for term in matrix[row_index * degree + col_index]:
+        monomials.append(polynomials.monomial(term, variables[col_index], 1))
+    for term in constant[row_index]:
+      monomials.append(polynomials.monomial(-term, None, 0))
+    equations.append(_make_equals_zero_split(monomials))
+
+  return equations
diff --git a/mathematics_dataset/mathematics_dataset/sample/linear_system_test.py b/mathematics_dataset/mathematics_dataset/sample/linear_system_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..5f4d24593b4760b43731904875be82a31ef1f396
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/linear_system_test.py
@@ -0,0 +1,56 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.sample.linear_system."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from absl.testing import absltest
+from absl.testing import parameterized
+from sample import linear_system
+from six.moves import range
+import sympy
+
+
+class ExpressionWithValueTest(parameterized.TestCase):
+
+  def testIsTrivialIn(self):
+    self.assertEqual(linear_system._is_trivial_in([[1]], 0), False)
+    self.assertEqual(linear_system._is_trivial_in([[1, 2], [3, 4]], 0), False)
+    self.assertEqual(linear_system._is_trivial_in([[1, 2], [3, 0]], 0), True)
+    self.assertEqual(linear_system._is_trivial_in([[1, 2], [3, 0]], 1), False)
+    self.assertEqual(linear_system._is_trivial_in([[1, 2], [0, 3]], 0), False)
+    self.assertEqual(linear_system._is_trivial_in([[1, 2], [0, 3]], 1), True)
+
+  @parameterized.parameters([1, 2, 3])
+  def testLinearSystem(self, degree):
+    for _ in range(100):  # test a few times
+      target = [random.randint(-100, 100) for _ in range(degree)]
+      variables = [sympy.Symbol(chr(ord('a') + i)) for i in range(degree)]
+      system = linear_system.linear_system(
+          variables=variables,
+          solutions=target,
+          entropy=10.0)
+      solved = sympy.solve(system, variables)
+      solved = [solved[symbol] for symbol in variables]
+      self.assertEqual(target, solved)
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/sample/number.py b/mathematics_dataset/mathematics_dataset/sample/number.py
new file mode 100644
index 0000000000000000000000000000000000000000..5f83dff880cb94fd67b265c95fb13dfeabe3058b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/number.py
@@ -0,0 +1,165 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Generate random integers and rationals with minimum guarantees on entropy."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import random
+
+# Dependency imports
+from util import display
+import numpy as np
+import six
+import sympy
+
+
+def _coprime_density(value):
+  """Returns float > 0; asymptotic density of integers coprime to `value`."""
+  factors = sympy.factorint(value)
+  density = 1.0
+  for prime in six.iterkeys(factors):
+    density *= 1 - 1 / prime
+  return density
+
+
+def integer(entropy, signed, min_abs=0, coprime_to=1):
+  """Returns an integer from a set of size ceil(10**entropy).
+
+  If `signed` is True, then includes negative integers, otherwise includes just
+  positive integers.
+
+  Args:
+    entropy: Float >= 0.
+    signed: Boolean. Whether to also return negative numbers.
+    min_abs: Integer >= 0. The minimum absolute value.
+    coprime_to: Optional integer >= 1. The returned integer is guaranteed to be
+        coprime to `coprime_to`, with entropy still accounted for.
+
+  Returns:
+    Integer.
+  """
+  assert isinstance(min_abs, int) and not isinstance(min_abs, bool)
+  coprime_to = abs(coprime_to)
+  assert min_abs >= 0
+
+  max_ = math.pow(10, entropy)
+  max_ += min_abs
+  if coprime_to >= 2:
+    max_ = max_ / _coprime_density(coprime_to) + 1
+
+  if signed:
+    max_ = int(math.ceil(max_ / 2))
+    range_ = [-max_, max_]
+  else:
+    max_ = int(math.ceil(max_))
+    range_ = [min_abs, max_]
+
+  while True:
+    value = random.randint(*range_)
+    if abs(value) >= min_abs and sympy.gcd(value, coprime_to) == 1:
+      break
+
+  return sympy.Integer(value)
+
+
+def non_integer_rational(entropy, signed):
+  """Similar args to `integer`. Entropy split between denom and numer."""
+  numer_entropy = random.uniform(0, entropy)
+  denom_entropy = entropy - numer_entropy
+  numer = integer(numer_entropy, signed, min_abs=1)
+  denom = integer(denom_entropy, False, min_abs=2, coprime_to=numer)
+  return sympy.Rational(numer, denom)
+
+
+def integer_or_rational(entropy, signed, min_abs=0):
+  """Returns a rational, with 50% probability of it being an integer."""
+  if random.choice([False, True]):
+    return integer(entropy, signed, min_abs=min_abs)
+  else:
+    return non_integer_rational(entropy, signed)
+
+
+def non_integer_decimal(entropy, signed):
+  """Returns a random decimal; integer divided by random power of ten.
+
+  Guaranteed to be non-integer (i.e., numbers after the decimal point).
+
+  Args:
+    entropy: Float.
+    signed: Boolean. Whether to also return negative numbers.
+
+  Returns:
+    Non-integer decimal.
+  """
+  while True:
+    base = integer(entropy, signed)
+    shift = random.randint(1, int(math.ceil(entropy)))
+    divisor = 10**shift
+    if base % divisor != 0:
+      return display.Decimal(sympy.Rational(base, divisor))
+
+
+def integer_or_decimal(entropy, signed):
+  """Returns integer or non-integer decimal; 50% probability of each."""
+  if random.choice([False, True]):
+    # Represent it as a decimal so that arithmetic operations are supported:
+    return display.Decimal(integer(entropy, signed))
+  else:
+    return non_integer_decimal(entropy, signed)
+
+
+def entropy_of_value(value):
+  """Returns "min entropy" that would give probability of getting this value."""
+  if isinstance(value, display.Decimal):
+    return entropy_of_value(sympy.numer(value))
+
+  if is_non_integer_rational(value):
+    numer = sympy.numer(value)
+    denom = sympy.denom(value)
+    return entropy_of_value(numer) + entropy_of_value(denom)
+  elif not is_integer(value):
+    raise ValueError('Unhandled value: {}'.format(value))
+
+  # Note: we sample integers in a range of size approx 10**entropy about zero,
+  # so assume that `abs(value)` is about half of the upper range.
+  return math.log10(5 * abs(value) + 1)
+
+
+def is_integer(value):
+  return isinstance(value, (int, np.int64, np.int32, sympy.Integer))
+
+
+def is_positive_integer(value):
+  """Filter for: value is a strictly positive integer."""
+  return is_integer(value) and value > 0
+
+
+def is_integer_or_rational(value):
+  return is_integer(value) or isinstance(value, sympy.Rational)
+
+
+def is_integer_or_decimal(value):
+  return is_integer(value) or isinstance(value, display.Decimal)
+
+
+def is_integer_or_rational_or_decimal(value):
+  return is_integer_or_rational(value) or is_integer_or_decimal(value)
+
+
+def is_non_integer_rational(value):
+  return is_integer_or_rational(value) and not is_integer(value)
diff --git a/mathematics_dataset/mathematics_dataset/sample/number_test.py b/mathematics_dataset/mathematics_dataset/sample/number_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..ab635e877206f1822d77e004e5b1941db313cd8f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/number_test.py
@@ -0,0 +1,129 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.sample.number."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from absl.testing import absltest
+from absl.testing import parameterized
+from sample import number
+from six.moves import range
+import sympy
+
+
+class NumberTest(parameterized.TestCase):
+
+  def testCoprimeDensity(self):
+    self.assertEqual(number._coprime_density(1), 1.0)
+    self.assertEqual(number._coprime_density(2), 0.5)
+    self.assertLess(abs(number._coprime_density(3) - 2/3), 1e-6)
+    self.assertLess(abs(number._coprime_density(6) - 1/3), 1e-6)
+
+  @parameterized.parameters(False, True)
+  def testInteger_allowZero(self, signed):
+    saw_zero = False
+    saw_nonzero = False
+    for _ in range(1000):
+      sample = number.integer(1, signed=signed)
+      if sample == 0:
+        saw_zero = True
+      else:
+        saw_nonzero = True
+      if saw_zero and saw_nonzero:
+        break
+    self.assertTrue(saw_zero)
+    self.assertTrue(saw_nonzero)
+
+  def testNonIntegerRational(self):
+    for _ in range(1000):
+      entropy = random.uniform(0, 10)
+      signed = random.choice([False, True])
+      sample = number.non_integer_rational(entropy, signed)
+      self.assertNotEqual(sympy.denom(sample), 1)
+
+  @parameterized.parameters(False, True)
+  def testIntegerOrRational(self, signed):
+    # Tests we can call it. Do it a few times so both code paths get executed.
+    for _ in range(10):
+      number.integer_or_rational(2, signed)
+
+  def testNonIntegerDecimal(self):
+    for _ in range(1000):
+      sample = number.non_integer_decimal(1, False)
+      self.assertNotEqual(sympy.denom(sample), 1)
+      self.assertLen(str(sample), 3)  # should be of form "0.n"
+      self.assertGreater(sample, 0)  # positive
+
+  def testNonIntegerDecimal_size(self):
+    saw_bigger_one = False
+    saw_smaller_one = False
+    for _ in range(1000):
+      sample = number.non_integer_decimal(2, False)
+      if sample > 1:
+        saw_bigger_one = True
+      else:
+        saw_smaller_one = True
+      if saw_bigger_one and saw_smaller_one:
+        break
+    self.assertTrue(saw_bigger_one)
+    self.assertTrue(saw_smaller_one)
+
+  @parameterized.parameters(
+      lambda: number.integer(0, True),
+      lambda: number.integer(1, True),
+      lambda: number.non_integer_rational(2, True),
+      lambda: number.non_integer_decimal(1, True))
+  def testGenerate_signed(self, generator):
+    saw_positive = False
+    saw_negative = False
+    for _ in range(1000):
+      sample = generator()
+      saw_positive |= sample > 0
+      saw_negative |= sample < 0
+      if saw_positive and saw_negative:
+        break
+
+    self.assertTrue(saw_positive)
+    self.assertTrue(saw_negative)
+
+  @parameterized.parameters(
+      lambda: number.integer(2, False),
+      lambda: number.non_integer_rational(2, False))
+  def testIntegerRational_distinctCount(self, generator):
+    seen = set()
+    for _ in range(3000):
+      seen.add(generator())
+    self.assertGreaterEqual(len(seen), 10 ** 2)
+
+  @parameterized.parameters(number.integer, number.non_integer_decimal)
+  def testEntropyOfValue(self, generator):
+    for entropy in [1, 2, 4, 8, 16]:
+      sum_entropy = 0.0
+      count = 2000
+      for _ in range(count):
+        value = generator(entropy, signed=True)
+        sum_entropy += number.entropy_of_value(value)
+      avg_entropy = sum_entropy / count
+      error = abs(entropy - avg_entropy) / entropy
+      self.assertLess(error, 0.2)
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/sample/ops.py b/mathematics_dataset/mathematics_dataset/sample/ops.py
new file mode 100644
index 0000000000000000000000000000000000000000..33c9738d4eec81ee0e18830746149b413dacc403
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/ops.py
@@ -0,0 +1,451 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Mathematical operations used to build up expressions for printing.
+
+We can't use sympy because sympy will automatically simplify many types of
+expressions, even with `evaluate=False` passed in. For example:
+
+*   Mul(-2, -3, evaluate=False) gives -(-6), not (-2) x (-3).
+*   Add(2, 1, evaluate=False) gives 1 + 2, because the terms are sorted.
+
+As such, it's easier just to work with our own op classes that display precisely
+as we created them. This also allows us to use custom symbols for the
+expressions, such as the multiplication symbol.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import abc
+
+# Dependency imports
+from absl import logging
+from sample import number
+from util import display
+import numpy as np
+import six
+from six.moves import zip
+import sympy
+
+
+MUL_SYMBOL = '*'
+DIV_SYMBOL = '/'
+POW_SYMBOL = '**'
+GT_SYMBOL = '>'
+LT_SYMBOL = '<'
+GE_SYMBOL = '>='
+LE_SYMBOL = '<='
+EQ_SYMBOL = '='
+NE_SYMBOL = '!='
+
+
+# Operator precedence levels. Used to insert brackets if necessary.
+_EQ_PRECEDENCE = 0
+_CONSTANT_PRECEDENCE = 1
+_POW_PRECEDENCE = 2
+_SQRT_PRECEDENCE = 3
+_MUL_PRECEDENCE = 4
+_ADD_PRECEDENCE = 5
+
+
+def bracketed(child, parent, bracket_if_same_precedence):
+  """Returns string representation of `child`, possibly bracketed.
+
+  Args:
+    child: Instance of `Op` or a valid value for `ConstantOp`.
+    parent: Instance of `Op`. Used to determine whether `child` needs to be
+        bracketed first before appearing in the parent op's expression.
+    bracket_if_same_precedence: Whether to bracket if the child has the same
+        operator precedence as the parent.
+
+  Returns:
+    String representation of `child`.
+  """
+  if not isinstance(child, Op):
+    child = Constant(child)
+
+  child_precedence = child.precedence
+  parent_precedence = parent.precedence
+  if (parent_precedence > child_precedence
+      or (parent_precedence == child_precedence
+          and not bracket_if_same_precedence)):
+    return str(child)
+  else:
+    return '({})'.format(child)
+
+
+def _flatten(iterable):
+  """Returns list."""
+  if isinstance(iterable, (list, tuple)):
+    result = list(iterable)
+  else:
+    assert isinstance(iterable, dict)
+    keys = sorted(six.iterkeys(iterable))
+    result = [iterable[key] for key in keys]
+  # Check we don't have any hierarchy in the structure (otherwise would need
+  # to use something recursive like tf.contrib.framework.nest.flatten).
+  for item in result:
+    assert not isinstance(item, (list, tuple, dict))
+  return result
+
+
+def _pack_sequence_as(example, flat):
+  if isinstance(example, list) or isinstance(example, tuple):
+    return flat
+  else:
+    assert isinstance(example, dict)
+    keys = sorted(six.iterkeys(example))
+    return {key: value for key, value in zip(keys, flat)}
+
+
+@six.add_metaclass(abc.ABCMeta)
+class Op(object):
+  """An operation.
+
+  This needs to support being transformed into sympy (and possibly in the future
+  other types such as an appropriately formatted string), when given the op
+  arguments.
+  """
+
+  def __init__(self, children):
+    """Initialize this `Op` base class.
+
+    Args:
+      children: Iterable structure containing child ops.
+    """
+    assert isinstance(children, (list, dict, tuple))
+    flat_children = _flatten(children)
+    flat_children = [child if isinstance(child, Op) else Constant(child)
+                     for child in flat_children]
+    children = _pack_sequence_as(children, flat_children)
+    self._children = children
+
+  @property
+  def children(self):
+    """Returns iterable or dict over immediate children."""
+    return self._children
+
+  def descendants(self):
+    """Returns list of all descendants (self, children, grandchildren, etc)."""
+    descendants = [self]
+    flat_children = _flatten(self._children)
+    for child in flat_children:
+      descendants += child.descendants()
+    return descendants
+
+  @abc.abstractmethod
+  def __str__(self):
+    """Returns a string format of this op."""
+
+  @abc.abstractmethod
+  def sympy(self):
+    """Returns the sympifcation of this op."""
+
+  def _sympy_(self):
+    """Convenience method to automatically sympify this object."""
+    try:
+      return self.sympy()
+    except AttributeError as e:
+      # Note: we print this error here, before raising it again, because sympy
+      # will think `AttributeError` refers to this object not having a `_sympy_`
+      # method, rather than having it, which leads to otherwise confusing error
+      # messages.
+      logging.error(
+          'Encountered attribute error while trying to sympify: %s', e)
+      raise e
+
+  @abc.abstractproperty
+  def precedence(self):
+    """Returns the precedence (integer) of this op."""
+
+
+class Constant(Op):
+  """Returns a constant value; a nullary op."""
+
+  def __init__(self, value):
+    super(Constant, self).__init__([])
+    if isinstance(value, six.integer_types):
+      value = sympy.Integer(value)
+    self._value = value
+
+  def __str__(self):
+    return str(self._value)
+
+  def sympy(self):
+    return self._value
+
+  @property
+  def value(self):
+    return self._value
+
+  @value.setter
+  def value(self, value):
+    self._value = value
+
+  def _is_simple(self):
+    """Returns whether it's a simple number, rather than a division or neg."""
+    if isinstance(self._value, sympy.Symbol):
+      return True
+    elif (isinstance(self._value, int)
+          or isinstance(self._value, sympy.Integer)
+          or isinstance(self._value, display.Decimal)
+          or isinstance(self._value, np.int64)
+          or isinstance(self._value, np.int32)):
+      return self._value >= 0
+    elif isinstance(self._value, sympy.Rational):
+      return False
+    elif isinstance(self._value, sympy.Function):
+      return True
+    else:
+      raise ValueError('Unknown type {}'.format(type(self._value)))
+
+  @property
+  def precedence(self):
+    if self._is_simple():
+      return _CONSTANT_PRECEDENCE
+    else:
+      return _MUL_PRECEDENCE
+
+
+class _SumLikeOp(Op):
+  """Abstract op for sum-like terms which may contain negative entries."""
+
+  @abc.abstractmethod
+  def expanded_signs_and_terms(self):
+    """Returns a list of arguments, plus any sub-arguments from sub-adds.
+
+    E.g., if this op is `Add(Add(2, Neg(3)), Mul(4, 5), 1)`, then will return
+    `[(True, 2), (False, 3), (True, Mul(4, 5)), (True, 1)]` (the arguments of
+    the inner add have been extracted).
+    """
+
+  def __str__(self):
+    signs_and_terms = self.expanded_signs_and_terms()
+    if not signs_and_terms:
+      return '0'
+    for i, (sign, term) in enumerate(signs_and_terms):
+      if i == 0:
+        if sign:
+          expression = bracketed(term, self, True)
+        else:
+          expression = '-' + bracketed(term, self, True)
+      else:
+        if sign:
+          expression += ' + ' + bracketed(term, self, True)
+        else:
+          expression += ' - ' + bracketed(term, self, True)
+    return expression
+
+
+class Identity(_SumLikeOp):
+  """The identity op (a unitary op)."""
+
+  def __init__(self, input_):
+    super(Identity, self).__init__({'input': input_})
+
+  def expanded_signs_and_terms(self):
+    if isinstance(self.children['input'], _SumLikeOp):
+      return self.children['input'].expanded_signs_and_terms()
+    else:
+      return [(True, self.children['input'])]
+
+  def __str__(self):
+    return str(self.children['input'])
+
+  def sympy(self):
+    return self.children['input'].sympy()
+
+  @property
+  def precedence(self):
+    return self.children['input'].precedence
+
+
+class Neg(_SumLikeOp):
+  """Negation, a unary op. Also has special display when appearing in a sum."""
+
+  def __init__(self, arg):
+    super(Neg, self).__init__({'input': arg})
+
+  def expanded_signs_and_terms(self):
+    if isinstance(self.children['input'], _SumLikeOp):
+      inner_signs_and_terms = self.children['input'].expanded_signs_and_terms()
+      return [(not sign, term) for (sign, term) in inner_signs_and_terms]
+    else:
+      return [(False, self.children['input'])]
+
+  def sympy(self):
+    return -sympy.sympify(self.children['input'])
+
+  def inner(self):
+    return self.children['input']
+
+  @property
+  def precedence(self):
+    return _ADD_PRECEDENCE
+
+
+class Add(_SumLikeOp):
+  """Addition."""
+
+  def __init__(self, *args):
+    super(Add, self).__init__(args)
+
+  def expanded_signs_and_terms(self):
+    """Returns a list of arguments, plus any sub-arguments from sub-adds.
+
+    E.g., if this op is `Add(Add(2, 3), Mul(4, 5), 1)`, then will return
+    `[2, 3, Mul(4, 5), 1]` (the arguments of the inner add have been extracted).
+    """
+    expanded = []
+    for arg in self.children:
+      if isinstance(arg, _SumLikeOp):
+        expanded += arg.expanded_signs_and_terms()
+      else:
+        expanded.append((True, arg))
+    return expanded
+
+  def sympy(self):
+    return sympy.Add(*[sympy.sympify(arg) for arg in self.children])
+
+  @property
+  def precedence(self):
+    return _ADD_PRECEDENCE
+
+
+class Sub(Op):
+  """Subtraction."""
+
+  def __init__(self, left, right):
+    super(Sub, self).__init__({'left': left, 'right': right})
+
+  def __str__(self):
+    return (bracketed(self.children['left'], self, False) + ' - '
+            + bracketed(self.children['right'], self, True))
+
+  def sympy(self):
+    return sympy.Add(
+        self.children['left'], sympy.Mul(-1, self.children['right']))
+
+  @property
+  def precedence(self):
+    return _ADD_PRECEDENCE
+
+
+class Mul(Op):
+  """Multiplication."""
+
+  def __init__(self, *args):
+    super(Mul, self).__init__(args)
+
+  def __str__(self):
+    if not self.children:
+      return '1'
+    else:
+      args = [bracketed(arg, self, False) for arg in self.children]
+      return MUL_SYMBOL.join(args)
+
+  def sympy(self):
+    return sympy.Mul(*[sympy.sympify(arg) for arg in self.children])
+
+  @property
+  def precedence(self):
+    return _MUL_PRECEDENCE
+
+
+class Div(Op):
+  """Division."""
+
+  def __init__(self, numer, denom):
+    super(Div, self).__init__({'numer': numer, 'denom': denom})
+
+  def __str__(self):
+    return u'{}{}{}'.format(
+        bracketed(self.children['numer'], self, True), DIV_SYMBOL,
+        bracketed(self.children['denom'], self, True))
+
+  def sympy(self):
+    return sympy.Mul(
+        self.children['numer'], sympy.Pow(self.children['denom'], -1))
+
+  @property
+  def precedence(self):
+    return _MUL_PRECEDENCE
+
+
+class Pow(Op):
+  """Power a to the power b."""
+
+  def __init__(self, a, b):
+    super(Pow, self).__init__({'a': a, 'b': b})
+
+  def __str__(self):
+    return u'{}{}{}'.format(
+        bracketed(self.children['a'], self, True), POW_SYMBOL,
+        bracketed(self.children['b'], self, True))
+
+  def sympy(self):
+    return sympy.Pow(
+        sympy.sympify(self.children['a']), sympy.sympify(self.children['b']))
+
+  @property
+  def precedence(self):
+    return _POW_PRECEDENCE
+
+
+class Sqrt(Op):
+  """Square root of a value."""
+
+  def __init__(self, a):
+    super(Sqrt, self).__init__({'a': a})
+
+  def __str__(self):
+    return 'sqrt({})'.format(self.children['a'])
+
+  def sympy(self):
+    return sympy.sqrt(self.children['a'])
+
+  @property
+  def precedence(self):
+    return _POW_PRECEDENCE
+
+
+class Eq(Op):
+  """Equality."""
+
+  def __init__(self, left, right):
+    super(Eq, self).__init__({'left': left, 'right': right})
+
+  def __str__(self):
+    return '{} = {}'.format(self.children['left'], self.children['right'])
+
+  def sympy(self):
+    return sympy.Eq(self.children['left'], self.children['right'])
+
+  @property
+  def precedence(self):
+    return _EQ_PRECEDENCE
+
+
+def number_constants(expressions):
+  """Returns list of integer, rational, decimal constants in the expressions."""
+  if isinstance(expressions, Op):
+    expressions = [expressions]
+  descendants = []
+  for expression in expressions:
+    descendants += expression.descendants()
+  candidate_constants = [op for op in descendants if isinstance(op, Constant)]
+  return [constant for constant in candidate_constants
+          if number.is_integer_or_rational_or_decimal(constant.value)]
diff --git a/mathematics_dataset/mathematics_dataset/sample/ops_test.py b/mathematics_dataset/mathematics_dataset/sample/ops_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..64837dc09e562611dfc52730e83d67dadb0011ba
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/ops_test.py
@@ -0,0 +1,206 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.sample.ops."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from sample import ops
+from six.moves import range
+import sympy
+
+
+class OpsTest(absltest.TestCase):
+
+  def testNeg(self):
+    op = ops.Neg(2)
+    self.assertEqual(str(op), '-2')
+    self.assertEqual(op.sympy(), -2)
+
+    op = ops.Add(ops.Neg(2), 3)
+    self.assertEqual(str(op), '-2 + 3')
+    self.assertEqual(op.sympy(), 1)
+
+    op = ops.Add(3, ops.Neg(2))
+    self.assertEqual(str(op), '3 - 2')
+    self.assertEqual(op.sympy(), 1)
+
+    op = ops.Add(ops.Add(ops.Neg(2), 5), 3)
+    self.assertEqual(str(op), '-2 + 5 + 3')
+    self.assertEqual(op.sympy(), 6)
+
+    op = ops.Add(3, ops.Add(ops.Identity(ops.Neg(2)), 5))
+    self.assertEqual(str(op), '3 - 2 + 5')
+    self.assertEqual(op.sympy(), 6)
+
+    op = ops.Add(3, ops.Add(2, ops.Neg(5)))
+    self.assertEqual(str(op), '3 + 2 - 5')
+    self.assertEqual(op.sympy(), 0)
+
+  def testAdd(self):
+    add = ops.Add()
+    self.assertEqual(str(add), '0')
+    self.assertEqual(add.sympy(), 0)
+
+    add = ops.Add(2, 3)
+    self.assertEqual(str(add), '2 + 3')
+    self.assertEqual(add.sympy(), 5)
+
+    add = ops.Add(ops.Add(1, 2), 3)
+    self.assertEqual(str(add), '1 + 2 + 3')
+    self.assertEqual(add.sympy(), 6)
+
+  def testSub(self):
+    sub = ops.Sub(2, 3)
+    self.assertEqual(str(sub), '2 - 3')
+    self.assertEqual(sub.sympy(), -1)
+
+    sub = ops.Sub(ops.Sub(1, 2), 3)
+    self.assertEqual(str(sub), '1 - 2 - 3')
+    self.assertEqual(sub.sympy(), -4)
+
+    sub = ops.Sub(1, ops.Sub(2, 3))
+    self.assertEqual(str(sub), '1 - (2 - 3)')
+    self.assertEqual(sub.sympy(), 2)
+
+    sub = ops.Sub(ops.Neg(1), 2)
+    self.assertEqual(str(sub), '-1 - 2')
+    self.assertEqual(sub.sympy(), -3)
+
+  def testMul(self):
+    mul = ops.Mul()
+    self.assertEqual(str(mul), '1')
+    self.assertEqual(mul.sympy(), 1)
+
+    mul = ops.Mul(2, 3)
+    self.assertEqual(str(mul), '2*3')
+    self.assertEqual(mul.sympy(), 6)
+
+    mul = ops.Mul(ops.Identity(ops.Constant(-2)), 3)
+    self.assertEqual(str(mul), '-2*3')
+    self.assertEqual(mul.sympy(), -6)
+
+    mul = ops.Mul(ops.Add(1, 2), 3)
+    self.assertEqual(str(mul), '(1 + 2)*3')
+    self.assertEqual(mul.sympy(), 9)
+
+    mul = ops.Mul(ops.Mul(2, 3), 5)
+    self.assertEqual(str(mul), '2*3*5')
+    self.assertEqual(mul.sympy(), 30)
+
+    # TODO(b/124038946): reconsider how we want brackets in these cases:
+#     mul = ops.Mul(ops.Div(2, 3), 5)
+#     self.assertEqual(str(mul), '(2/3)*5')
+#     self.assertEqual(mul.sympy(), sympy.Rational(10, 3))
+#
+#     mul = ops.Mul(sympy.Rational(2, 3), 5)
+#     self.assertEqual(str(mul), '(2/3)*5')
+#     self.assertEqual(mul.sympy(), sympy.Rational(10, 3))
+
+  def testDiv(self):
+    div = ops.Div(2, 3)
+    self.assertEqual(str(div), '2/3')
+    self.assertEqual(div.sympy(), sympy.Rational(2, 3))
+
+    div = ops.Div(2, sympy.Rational(4, 5))
+    self.assertEqual(str(div), '2/(4/5)')
+    self.assertEqual(div.sympy(), sympy.Rational(5, 2))
+
+    div = ops.Div(1, ops.Div(2, 3))
+    self.assertEqual(str(div), '1/(2/3)')
+    self.assertEqual(div.sympy(), sympy.Rational(3, 2))
+
+    div = ops.Div(ops.Div(2, 3), 4)
+    self.assertEqual(str(div), '(2/3)/4')
+    self.assertEqual(div.sympy(), sympy.Rational(1, 6))
+
+    div = ops.Div(2, ops.Mul(3, 4))
+    self.assertEqual(str(div), '2/(3*4)')
+
+    div = ops.Div(2, sympy.Function('f')(sympy.Symbol('x')))
+    self.assertEqual(str(div), '2/f(x)')
+
+  def testPow(self):
+    pow_ = ops.Pow(2, 3)
+    self.assertEqual(str(pow_), '2**3')
+    self.assertEqual(pow_.sympy(), 8)
+
+    pow_ = ops.Pow(4, sympy.Rational(1, 2))
+    self.assertEqual(str(pow_), '4**(1/2)')
+    self.assertEqual(pow_.sympy(), 2)
+
+    pow_ = ops.Pow(sympy.Rational(1, 2), 3)
+    self.assertEqual(str(pow_), '(1/2)**3')
+    self.assertEqual(pow_.sympy(), 1/8)
+
+    pow_ = ops.Pow(3, ops.Pow(2, 1))
+    self.assertEqual(str(pow_), '3**(2**1)')
+    self.assertEqual(pow_.sympy(), 9)
+
+    pow_ = ops.Pow(ops.Pow(2, 3), 4)
+    self.assertEqual(str(pow_), '(2**3)**4')
+    self.assertEqual(pow_.sympy(), 4096)
+
+    pow_ = ops.Pow(-5, 2)
+    self.assertEqual(str(pow_), '(-5)**2')
+    self.assertEqual(pow_.sympy(), 25)
+
+  def testEq(self):
+    op = ops.Eq(ops.Add(2, 3), 4)
+    self.assertEqual(str(op), '2 + 3 = 4')
+    self.assertEqual(op.sympy(), False)
+
+  def testDescendants(self):
+    constants = [ops.Constant(i) for i in range(6)]
+
+    # (1 + 2*3**4) / 5 - 6
+    expression = ops.Sub(
+        ops.Div(
+            ops.Add(
+                constants[0],
+                ops.Mul(
+                    constants[1],
+                    ops.Pow(
+                        constants[2],
+                        constants[3]))),
+            constants[4]),
+        constants[5])
+    descendants = expression.descendants()
+    descendants = ops._flatten(descendants)
+
+    for constant in constants:
+      self.assertIn(constant, descendants)
+      self.assertEqual(descendants.count(constant), 1)
+
+    # Also test top-level.
+    self.assertEqual(constants[0].descendants(), [constants[0]])
+
+    # Also general structure.
+    constant = ops.Constant(3)
+    expression = ops.Neg(constant)
+    self.assertEqual(set(expression.descendants()), set([constant, expression]))
+
+  def testNumberConstants(self):
+    constant = ops.Constant(3)
+    expression = ops.Neg(constant)
+    constants = ops.number_constants([expression])
+    self.assertEqual(constants, [constant])
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/sample/polynomials.py b/mathematics_dataset/mathematics_dataset/sample/polynomials.py
new file mode 100644
index 0000000000000000000000000000000000000000..09ee604b5ee133a31afd2402614a00c7bda297cf
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/polynomials.py
@@ -0,0 +1,601 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Generate polynomials with given values."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import random
+
+# Dependency imports
+from sample import number
+from sample import ops
+from util import combinatorics
+import numpy as np
+import six
+from six.moves import range
+from six.moves import zip
+import sympy
+from sympy.solvers.diophantine.diophantine import base_solution_linear as diophantine_solve_linear_2d
+
+
+def expanded_coefficient_counts(length, is_zero):
+  """Generates list of integers for number of terms of given power.
+
+  Args:
+    length: Integer >= `sum(is_zero)`.
+    is_zero: List of booleans.
+
+  Returns:
+    List of non-negative integers of length `is_zero`, summing to `length`,
+    such that if `is_zero[i]` then `return_value[i] != 1`.
+
+  Raises:
+    ValueError: If assignment not possible.
+  """
+  if length == 1 and all(is_zero):
+    raise ValueError('length=1 and all zero')
+
+  counts = np.asarray([0 if zero else 1 for zero in is_zero])
+  extra_needed = (length - sum(counts))
+
+  if extra_needed < 0:
+    raise ValueError('length={} cannot handle is_zero={}'
+                     .format(length, is_zero))
+
+  extra = combinatorics.uniform_non_negative_integers_with_sum(
+      count=len(is_zero), sum_=extra_needed)
+  counts += np.asarray(extra)
+
+  # Tweak so that no zeros get "1".
+  while True:
+    bad_zeros = [
+        i for i in range(len(is_zero)) if is_zero[i] and counts[i] == 1
+    ]
+    if not bad_zeros:
+      break
+    take_from = random.choice(bad_zeros)
+    add_to = random.choice(
+        [i for i in range(len(is_zero)) if counts[i] >= 1 and i != take_from])
+    counts[take_from] -= 1
+    counts[add_to] += 1
+
+  return counts
+
+
+def _split_value_equally(delta, count):
+  """Splits an integer or rational into roughly equal parts."""
+  numer = sympy.numer(delta)
+  denom = sympy.denom(delta)
+  return [int(math.floor((numer + i) / count)) / denom for i in range(count)]
+
+
+def integers_with_sum(value, count, entropy):
+  """Returns list of integers with a given sum.
+
+  Args:
+    value: Target value.
+    count: Integer >= 1; the number of integers to use.
+    entropy: Entropy to use (in total).
+
+  Returns:
+    List of numbers summing to `value`.
+
+  Raises:
+    ValueError: If `value` is not an integer.
+  """
+  # Special cases.
+  if count == 0:
+    assert value == 0
+    assert entropy == 0
+    return []
+  if count == 1:
+    assert entropy == 0
+    return [value]
+
+  if not number.is_integer(value):
+    raise ValueError('value={} (type={}) is not an integer'
+                     .format(value, type(value)))
+
+  # Because e.g., (1, 1) and (2, 2) will both map to the same set of integers
+  # when we normalize to have sum equal to `value`.
+  entropy *= count / (count - 1)
+
+  min_term_entropy = max(
+      1, number.entropy_of_value(int(math.ceil(value/count))))
+  term_entropies = entropy * np.random.dirichlet(np.ones(count))
+  term_entropies = np.maximum(min_term_entropy, term_entropies)
+
+  terms = [number.integer(term_entropy, signed=True)
+           for term_entropy in term_entropies]
+
+  delta = value - sum(terms)
+  deltas = _split_value_equally(delta, count)
+  terms = [term + delta for term, delta in zip(terms, deltas)]
+  random.shuffle(terms)
+  return terms
+
+
+def monomial(coefficient, variables, powers):
+  """Makes a simple monomial term."""
+  if not isinstance(variables, (list, tuple)):
+    variables = [variables]
+  if not isinstance(powers, (list, tuple, np.ndarray)):
+    powers = [powers]
+
+  terms = []
+
+  for variable, power in zip(variables, powers):
+    if power == 0:
+      continue
+    elif power == 1:
+      terms.append(variable)
+    else:
+      terms.append(ops.Pow(variable, power))
+
+  if (not terms
+      or isinstance(coefficient, sympy.Symbol)
+      or abs(coefficient) != 1):
+    if isinstance(coefficient, sympy.Symbol):
+      terms.insert(0, coefficient)
+    else:
+      terms.insert(0, abs(coefficient))
+
+  if len(terms) > 1:
+    term = ops.Mul(*terms)
+  else:
+    term = terms[0]
+
+  if not isinstance(coefficient, sympy.Symbol) and coefficient < 0:
+    term = ops.Neg(term)
+
+  return term
+
+
+def sample_coefficients(degrees, entropy, min_non_zero=0, max_non_zero=None):
+  """Generates grid of coefficients with shape `degrees + 1`.
+
+  This corresponds to univariate if degrees has length 1, otherwise
+  multivariate.
+
+  Args:
+    degrees: List of integers containing max degrees of variables.
+    entropy: Float >= 0; entropy for generating entries.
+    min_non_zero: Optional integer >= 1; the minimum number of non-zero coeffs.
+    max_non_zero: Optional integer >= 1; the maximum number of non-zero coeffs.
+
+  Returns:
+    NumPy int array of shape `degrees + 1`.
+  """
+  if isinstance(degrees, int):
+    degrees = [degrees]
+  degrees = np.asarray(degrees)
+
+  def random_index():
+    return [random.randint(0, degrees[i]) for i in range(len(degrees))]
+
+  indices = set()
+  # Ensure a variable of degree `degrees[i]` occurs for every axis i.
+  for i, degree in enumerate(degrees):
+    if degree > 0:
+      index = random_index()
+      index[i] = degree
+      indices.add(tuple(index))
+
+  abs_max_non_zero = np.prod(degrees + 1)
+
+  min_non_zero = max(min_non_zero, 1, len(indices))
+  if max_non_zero is None:
+    max_non_zero = min_non_zero + int(entropy/2)
+
+  min_non_zero = min(min_non_zero, abs_max_non_zero)
+  max_non_zero = min(max_non_zero, abs_max_non_zero)
+  max_non_zero = max(min_non_zero, max_non_zero)
+
+  num_non_zero = random.randint(min_non_zero, max_non_zero)
+
+  while len(indices) < num_non_zero:
+    indices.add(tuple(random_index()))
+
+  coeffs = np.zeros(degrees + 1, dtype=np.int64)
+  entropies = entropy * np.random.dirichlet(np.ones(num_non_zero))
+
+  for index, entry_entropy in zip(indices, entropies):
+    value = number.integer(entry_entropy, signed=True, min_abs=1)
+    coeffs.itemset(index, value)
+
+  return coeffs
+
+
+def expand_coefficients(coefficients, entropy, length=None):
+  """Expands coefficients to multiple terms that sum to each coefficient.
+
+  Args:
+    coefficients: Array, such that `coefficients[i, j, ..., k]` is the
+        coefficient of x**i * y**j * ... * z**k.
+    entropy: Float >= 0; the entropy to use for generating extra randomness.
+    length: Number of terms that appear, e.g., 2x + 3 has two terms. If `None`
+        then a suitable length will be picked depending on the entropy
+        requested.
+
+  Returns:
+    Numpy object array with the same shape as `coefficients`, containing lists.
+  """
+  coefficients = np.asarray(coefficients)
+  shape = coefficients.shape
+
+  expanded_coefficients = np.empty(shape, dtype=np.object_ )
+
+  min_length = np.count_nonzero(coefficients) + 2
+  if length is None:
+    max_length = min_length + int(math.ceil(entropy) / 2)
+    length = random.randint(min_length, max_length)
+  if length < min_length:
+    length = min_length
+
+  is_zero_flat = np.reshape(coefficients, [-1]) == 0
+  counts = expanded_coefficient_counts(length, is_zero=is_zero_flat)
+  coeffs_entropy = entropy * np.random.dirichlet(np.maximum(1e-9, counts - 1))
+  counts = np.reshape(counts, shape)
+  coeffs_entropy = np.reshape(coeffs_entropy, shape)
+
+  indices = list(zip(*np.indices(shape).reshape([len(shape), -1])))
+  for power in indices:
+    coeffs = integers_with_sum(
+        value=coefficients.item(power),
+        count=counts.item(power),
+        entropy=coeffs_entropy.item(power))
+    expanded_coefficients.itemset(power, coeffs)
+
+  return expanded_coefficients
+
+
+def sample_expanded_coefficients(degrees, entropy, length=None):
+  """Convenience function: samples and expands coeffs, entropy split equally."""
+  coefficients = sample_coefficients(degrees, entropy/2, max_non_zero=length)
+  return expand_coefficients(coefficients, entropy/2, length)
+
+
+def coefficients_to_polynomial(coefficients, variables):
+  """Converts array of lists of coefficients to a polynomial."""
+  coefficients = np.asarray(coefficients)
+  shape = coefficients.shape
+
+  indices = list(zip(*np.indices(shape).reshape([len(shape), -1])))
+  monomials = []
+  for power in indices:
+    coeffs = coefficients.item(power)
+    if (number.is_integer_or_rational(coeffs)
+        or isinstance(coeffs, sympy.Symbol)):
+      coeffs = [coeffs]
+    elif not isinstance(coeffs, list):
+      raise ValueError('Unrecognized coeffs={} type={}'
+                       .format(coeffs, type(coeffs)))
+    for coeff in coeffs:
+      monomials.append(monomial(coeff, variables, power))
+  random.shuffle(monomials)
+  return ops.Add(*monomials)
+
+
+def sample(variables, degrees, entropy, length=None):
+  coefficients = sample_expanded_coefficients(degrees, entropy, length)
+  return coefficients_to_polynomial(coefficients, variables)
+
+
+def add_coefficients(coeffs1, coeffs2):
+  """Adds together two sets of coefficients over same set of variables."""
+  coeffs1 = np.asarray(coeffs1)
+  coeffs2 = np.asarray(coeffs2)
+
+  degrees1 = np.array(coeffs1.shape)
+  degrees2 = np.array(coeffs2.shape)
+  assert len(degrees1) == len(degrees2)
+
+  extra1 = np.maximum(0, degrees2 - degrees1)
+  extra2 = np.maximum(0, degrees1 - degrees2)
+
+  pad1 = [(0, extra) for extra in extra1]
+  pad2 = [(0, extra) for extra in extra2]
+
+  coeffs1 = np.pad(coeffs1, pad1, 'constant', constant_values=0)
+  coeffs2 = np.pad(coeffs2, pad2, 'constant', constant_values=0)
+
+  return coeffs1 + coeffs2
+
+
+def _random_factor(integer):
+  factors = sympy.factorint(integer)
+  result = 1
+  for factor, power in six.iteritems(factors):
+    result *= factor ** random.randint(0, power)
+  return result
+
+
+def coefficients_linear_split(coefficients, entropy):
+  """Finds two sets of coefficients and multipliers summing to `coefficients`.
+
+  Given `coefficients` (an integer vector), will sample integers `a, b`, and
+  two sets of coefficients `coefficients_1, coefficients_2`, such that
+  `a * coefficients_1 + b * coefficients_2 == coefficients`.
+
+  Args:
+    coefficients: Array of coefficients.
+    entropy: Float >= 0; the amount of randomness used to sample.
+
+  Returns:
+    Tuple (a, b, coefficients_1, coefficients_2)`.
+  """
+  coefficients = np.asarray(coefficients)
+  coefficients_shape = coefficients.shape
+  coefficients = np.reshape(coefficients, [-1])
+
+  entropy_a = max(1, random.uniform(0, entropy/3))
+  entropy_b = max(1, random.uniform(0, entropy/3))
+  entropy -= entropy_a + entropy_b
+  entropy_coefficients = entropy * np.random.dirichlet(
+      np.ones(len(coefficients)))
+
+  # For each target coefficient z, we are required to solve the linear
+  # Diophantine equation a*x + b*y = c. Bezout's theorem: this has a solution if
+  # and only if gcd(a, b) divides c.
+  # Thus to be solvable for all coefficients, a and b must be chosen such that
+  # gcd(a, b) divides the gcd of the coefficients.
+  coefficients_gcd = sympy.gcd([i for i in coefficients])
+  coefficients_gcd = max(1, abs(coefficients_gcd))
+
+  a = number.integer(entropy_a, signed=True, min_abs=1)
+  b = number.integer(entropy_b, signed=True, min_abs=1, coprime_to=a)
+  b *= _random_factor(coefficients_gcd)
+  if random.choice([False, True]):
+    a, b = b, a
+
+  coefficients_1 = np.zeros(coefficients.shape, dtype=np.object_)
+  coefficients_2 = np.zeros(coefficients.shape, dtype=np.object_)
+
+  for index, coefficient in enumerate(coefficients):
+    entropy_coeff = entropy_coefficients[index]
+    t = number.integer(entropy_coeff, signed=True)
+    x, y = diophantine_solve_linear_2d(c=coefficient, a=a, b=b, t=t)
+    coefficients_1[index] = x
+    coefficients_2[index] = y
+
+  # Prevent all coefficients from being zero.
+  while np.all(coefficients_1 == 0) or np.all(coefficients_2 == 0):
+    index = random.randint(0, len(coefficients) - 1)
+    scale = random.choice([-1, 1])
+    coefficients_1[index] += scale * b
+    coefficients_2[index] -= scale * a
+
+  coefficients_1 = np.reshape(coefficients_1, coefficients_shape)
+  coefficients_2 = np.reshape(coefficients_2, coefficients_shape)
+
+  return a, b, coefficients_1, coefficients_2
+
+
+def _degree_of_variable(polynomial, variable):
+  polynomial = sympy.sympify(polynomial).expand()
+  if polynomial.is_constant():
+    return 0
+  polynomial = sympy.poly(polynomial)
+  if variable not in polynomial.free_symbols:
+    return 0
+  return polynomial.degree(variable)
+
+
+def _sample_with_brackets(depth, variables, degrees, entropy, length,
+                          force_brackets=True):
+  """Internal recursive function for: constructs a polynomial with brackets."""
+  # To generate arbitrary polynomial recursively, can do one of:
+  # *   add two polynomials, with at least one having brackets.
+  # *   multiply two polynomials.
+  # *   call `sample` (i.e., polynomial without brackets).
+
+  if force_brackets:
+    length = max(2, length)
+
+  if not force_brackets and (random.choice([False, True]) or length < 2):
+    return sample(variables, degrees, entropy, length)
+
+  length_left = random.randint(1, length - 1)
+  length_right = length - length_left
+  entropy_left, entropy_right = entropy * np.random.dirichlet(
+      [length_left, length_right])
+
+  if random.choice([False, True]):
+    # Add two. Force brackets on at least one of the polynomials, and sample
+    # repeatedly until we don't get cancellation.
+    while True:
+      left = _sample_with_brackets(
+          depth + 1, variables, degrees, entropy_left, length_left, True)
+      right = _sample_with_brackets(
+          depth + 1, variables, degrees, entropy_right, length_right, False)
+      if random.choice([False, True]):
+        left, right = right, left
+      result = ops.Add(left, right)
+      all_ok = True
+      for variable, degree in zip(variables, degrees):
+        if _degree_of_variable(result, variable) != degree:
+          all_ok = False
+          break
+      if all_ok:
+        return result
+  else:
+    # Multiply two.
+    def sample_with_zero_check(degrees_, entropy_, length_):
+      while True:
+        result = _sample_with_brackets(
+            depth + 1, variables, degrees_, entropy_, length_, False)
+        if degrees_.sum() > 0 or not result.sympy().is_zero:
+          return result
+    degrees = np.asarray(degrees)
+
+    def sample_degree(max_degree):
+      """Select in range [0, max_degree], biased away from ends."""
+      if max_degree <= 1 or random.choice([False, True]):
+        return random.randint(0, max_degree)
+      return random.randint(1, max_degree - 1)
+
+    degrees_left = np.array([sample_degree(degree) for degree in degrees])
+    degrees_right = degrees - degrees_left
+    left = sample_with_zero_check(degrees_left, entropy_left, length_left)
+    right = sample_with_zero_check(degrees_right, entropy_right, length_right)
+    return ops.Mul(left, right)
+
+
+def sample_with_brackets(variables, degrees, entropy, length=None):
+  """Constructs a polynomial with brackets.
+
+  Args:
+    variables: List of variables to use.
+    degrees: Max degrees of variables. This function guarantees that these will
+        be obtained in the returned polynomial.
+    entropy: Float >= 0; the randomness to use in generating the polynomial.
+    length: Optional integer containing number of terms. If `None` then an
+        appropriate one will be generated depending on the entropy.
+
+  Returns:
+    Instance of `ops.Op` containing the polynomial.
+  """
+  if isinstance(degrees, int):
+    degrees = [degrees]
+  if not isinstance(variables, (list, tuple)):
+    variables = [variables]
+
+  if length is None:
+    length = 3 + random.randint(0, int(entropy/2))
+
+  # Add on some entropy to compensate for different expressions generating the
+  # same apparent polynomial.
+  entropy += combinatorics.log_number_binary_trees(length) / math.log(10)
+
+  return _sample_with_brackets(0, variables, degrees, entropy, length, True)
+
+
+def sample_with_small_evaluation(variable, degree, max_abs_input, entropy):
+  """Generates a (canonically ordered) polynomial, with bounded evaluation.
+
+  The coefficients are chosen to make use of the entropy, with the scaling
+  adjusted so that all give roughly the same contribution to the output of the
+  polynomial when the input is bounded in magnitude by `max_abs_input`.
+
+  Args:
+    variable: Variable to use in polynomial.
+    degree: Degree of polynomial.
+    max_abs_input: Number >= 1; max absolute value of input.
+    entropy: Float; randomness for generating polynomial.
+
+  Returns:
+    Instance of `ops.Add`.
+  """
+  assert max_abs_input >= 1
+  entropies = entropy * np.random.dirichlet(np.ones(degree + 1))
+  coeffs = []
+
+  for power in range(degree + 1):
+    # This scaling guarantees that the terms give roughly equal contribution
+    # to the typical magnitude of the polynomial when |input| <= max_abs_input.
+    delta = 0.5 * (degree - 2 * power) * math.log10(max_abs_input)
+    power_entropy = entropies[power] + delta
+    min_abs = 1 if power == degree else 0
+    coeff = number.integer(power_entropy, signed=True, min_abs=min_abs)
+    coeffs.append(coeff)
+
+  terms = [monomial(coeff, variable, power)
+           for power, coeff in enumerate(coeffs)]
+  return ops.Add(*terms)
+
+
+def sample_messy_power(variable, entropy):
+  """Returns unsimplified power expression like ((x**2)**3/x**4)**2/x**3."""
+  if entropy <= 0:
+    return variable
+
+  which = random.choice([1, 2, 3])
+
+  if which == 1:
+    exponent_entropy = min(2, entropy)
+    entropy -= exponent_entropy
+    exponent = number.integer_or_rational(exponent_entropy, signed=True)
+    left = sample_messy_power(variable, entropy)
+    return ops.Pow(left, exponent)
+
+  entropy_left = entropy / 2
+  if entropy_left < 1:
+    entropy_left = 0
+  entropy_right = entropy - entropy_left
+  if random.choice([False, True]):
+    entropy_left, entropy_right = entropy_right, entropy_left
+
+  left = sample_messy_power(variable, entropy_left)
+  right = sample_messy_power(variable, entropy_right)
+  if which == 2:
+    return ops.Mul(left, right)
+  else:
+    return ops.Div(left, right)
+
+
+def trim(coefficients):
+  """Makes non-zero entry in the final slice along each axis."""
+  coefficients = np.asarray(coefficients)
+  non_zero = np.not_equal(coefficients, 0)
+  ndim = coefficients.ndim
+
+  for axis in range(ndim):
+    length = coefficients.shape[axis]
+    axis_complement = list(range(0, axis)) + list(range(axis + 1, ndim))
+    non_zero_along_axis = np.any(non_zero, axis=tuple(axis_complement))
+
+    slice_to = 0
+    for index in range(length - 1, -1, -1):
+      if non_zero_along_axis[index]:
+        slice_to = index + 1
+        break
+
+    if slice_to < length:
+      coefficients = coefficients.take(axis=axis, indices=list(range(slice_to)))
+
+  return coefficients
+
+
+def differentiate(coefficients, axis):
+  """Differentiate coefficients (corresponding to polynomial) along axis."""
+  coefficients = np.asarray(coefficients)
+  indices = list(range(1, coefficients.shape[axis]))
+  coefficients = coefficients.take(axis=axis, indices=indices)
+
+  broadcast_shape = np.ones(coefficients.ndim, dtype=np.int32)
+  broadcast_shape[axis] = len(indices)
+  broadcast = np.asarray(indices).reshape(broadcast_shape)
+
+  result = broadcast * coefficients
+  return trim(result)
+
+
+def integrate(coefficients, axis):
+  """Integrate coefficients (corresponding to polynomial) along axis."""
+  coefficients = np.asarray(coefficients)
+
+  length = coefficients.shape[axis]
+  broadcast_shape = np.ones(coefficients.ndim, dtype=np.int32)
+  broadcast_shape[axis] = length
+  powers = np.array([sympy.Integer(i) for i in range(1, length + 1)])
+  powers = powers.reshape(broadcast_shape)
+
+  result_unpadded = coefficients / powers
+
+  pad = [(1 if i == axis else 0, 0) for i in range(coefficients.ndim)]
+  return np.pad(result_unpadded, pad, 'constant', constant_values=0)
diff --git a/mathematics_dataset/mathematics_dataset/sample/polynomials_test.py b/mathematics_dataset/mathematics_dataset/sample/polynomials_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..c2879e1ecc47cf9cf26c5a34fe1f68b5a67e604a
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/sample/polynomials_test.py
@@ -0,0 +1,162 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.sample.polynomials."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import random
+
+# Dependency imports
+from absl.testing import parameterized
+from sample import polynomials
+import numpy as np
+from six.moves import range
+import sympy
+import tensorflow as tf
+
+
+class ExpressionWithValueTest(tf.test.TestCase, parameterized.TestCase):
+
+  def testSplitValueEqually(self):
+    split = polynomials._split_value_equally(3, 2)
+    self.assertEqual(split, [1, 2])
+    split = polynomials._split_value_equally(sympy.sympify('3/4'), 2)
+    self.assertEqual(split, [sympy.sympify('1/4'), sympy.sympify('1/2')])
+
+  def testIntegersWithSum(self):
+    value = 13
+    count = 10
+    terms = polynomials.integers_with_sum(value=value, count=count, entropy=4.0)
+    self.assertLen(terms, count)
+    self.assertEqual(sum(terms), value)
+
+  def testMonomial(self):
+    x, y = sympy.symbols('x y')
+    self.assertEqual(str(polynomials.monomial(1, [x, y], [2, 3])), 'x**2*y**3')
+    # TODO(b/124038530): how handle rational coefficients; are they even used?
+    # self.assertEqual(
+    #     str(polynomials.monomial(sympy.Rational(2, 3), [x], [1])), '2*x/3')
+    # self.assertEqual(
+    #     str(polynomials.monomial(sympy.Rational(1, 3), [x], [1])), 'x/3')
+    self.assertEqual(str(polynomials.monomial(x, [y], [4])), 'x*y**4')
+
+  def testExpandCoefficients(self):
+    for _ in range(10):
+      num_variables = np.random.randint(1, 4)
+      degrees = np.random.randint(0, 4, [num_variables])
+      coefficients = np.random.randint(-3, 3, degrees + 1)
+      entropy = np.random.uniform(0, 10)
+      expanded = polynomials.expand_coefficients(coefficients, entropy)
+      collapsed = np.vectorize(sum)(expanded)
+      self.assertAllEqual(coefficients, collapsed)
+
+  def testCoefficientsToPolynomial(self):
+    coeffs = [3, 2, 1]
+    x = sympy.Symbol('x')
+    polynomial = polynomials.coefficients_to_polynomial(coeffs, [x])
+    polynomial = sympy.sympify(polynomial)
+    self.assertEqual(polynomial, x*x + 2*x + 3)
+
+  def testUnivariate(self):
+    # Test generation for: x**2 + 2*x + 1
+    x = sympy.Symbol('x')
+    coeffs = [1, 2, 3]
+    for _ in range(10):
+      expanded = polynomials.expand_coefficients(coeffs, 5.0)
+      polynomial = polynomials.coefficients_to_polynomial(expanded, [x])
+      sympified = sympy.sympify(polynomial)
+      self.assertEqual(sympified, 1 + 2*x + 3*x*x)
+
+  def testMultivariate(self):
+    # Test generation for: x**2 + 2*x*y + 3*y**2 - x + 5
+    x, y = sympy.symbols('x y')
+    coeffs = [[5, 0, 3], [-1, 2, 0], [1, 0, 0]]
+    for _ in range(10):
+      expanded = polynomials.expand_coefficients(coeffs, 5.0, length=10)
+      polynomial = polynomials.coefficients_to_polynomial(expanded, [x, y])
+      sympified = sympy.sympify(polynomial)
+      self.assertEqual(sympified, x*x + 2*x*y + 3*y*y - x + 5)
+
+  def testAddCoefficients(self):
+    # Add x**2 + 2*y and 3*x + 4*y**3.
+    coeffs1 = [[0, 2], [0, 0], [1, 0]]
+    coeffs2 = [[0, 0, 0, 4], [3, 0, 0, 0]]
+    target = [[0, 2, 0, 4], [3, 0, 0, 0], [1, 0, 0, 0]]
+    actual = polynomials.add_coefficients(coeffs1, coeffs2)
+    self.assertAllEqual(target, actual)
+
+  def testCoefficientsLinearSplit(self):
+    for degree in range(3):
+      for ndims in range(3):
+        for _ in range(10):
+          coefficients = np.random.randint(-5, 5, [degree + 1] * ndims)
+          entropy = random.uniform(1, 4)
+          c1, c2, coeffs1, coeffs2 = polynomials.coefficients_linear_split(
+              coefficients, entropy)
+          c1 = int(c1)
+          c2 = int(c2)
+          coeffs1 = np.asarray(coeffs1, dtype=np.int32)
+          coeffs2 = np.asarray(coeffs2, dtype=np.int32)
+          sum_ = c1 * coeffs1 + c2 * coeffs2
+          self.assertAllEqual(sum_, coefficients)
+
+  def testSampleWithBrackets(self):
+    x, y = sympy.symbols('x y')
+    for _ in range(100):
+      degrees = np.random.randint(1, 4, [2])
+      entropy = random.uniform(0, 4)
+      polynomial = polynomials.sample_with_brackets(
+          variables=[x, y], degrees=degrees, entropy=entropy)
+      self.assertIn('(', str(polynomial))
+      poly = sympy.poly(sympy.sympify(polynomial).expand())
+      self.assertEqual(poly.degree(x), degrees[0])
+      self.assertEqual(poly.degree(y), degrees[1])
+
+  def testTrim(self):
+    self.assertAllEqual(polynomials.trim([1]), [1])
+    self.assertAllEqual(polynomials.trim([1, 0]), [1])
+    self.assertAllEqual(polynomials.trim([0, 1]), [0, 1])
+    self.assertAllEqual(polynomials.trim([0]), [])
+    self.assertAllEqual(polynomials.trim([0, 0]), [])
+
+  def testDifferentiate_univariate(self):
+    coeffs = [5, 3, 2]
+    expected = [3, 4]
+    actual = polynomials.differentiate(coeffs, 0)
+    self.assertAllEqual(expected, actual)
+
+  def testDifferentiate_multivariate(self):
+    coeffs = [[0, 3, 1], [5, 0, 0], [0, 2, 0]]
+    expected = [[5, 0], [0, 4]]
+    actual = polynomials.differentiate(coeffs, 0)
+    self.assertAllEqual(expected, actual)
+
+  def testIntegrate_univariate(self):
+    coeffs = [5, 3, 2]
+    expected = [0, 5, sympy.Rational(3, 2), sympy.Rational(2, 3)]
+    actual = polynomials.integrate(coeffs, 0)
+    self.assertAllEqual(expected, actual)
+
+  def testIntegrate_multivariate(self):
+    coeffs = [[0, 1], [1, 0]]
+    expected = [[0, 0, sympy.Rational(1, 2)], [0, 1, 0]]
+    actual = polynomials.integrate(coeffs, 1)
+    self.assertAllEqual(expected, actual)
+
+
+if __name__ == '__main__':
+  tf.test.main()
diff --git a/mathematics_dataset/mathematics_dataset/util/__init__.py b/mathematics_dataset/mathematics_dataset/util/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..d780829a1f5101333577a1ca9adcb960529213f1
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/__init__.py
@@ -0,0 +1,13 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/mathematics_dataset/mathematics_dataset/util/combinatorics.py b/mathematics_dataset/mathematics_dataset/util/combinatorics.py
new file mode 100644
index 0000000000000000000000000000000000000000..92eb8288622c3fd9b6f9516619674e1b7db1c082
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/combinatorics.py
@@ -0,0 +1,58 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Combinatorics utility functions."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import random
+
+# Dependency imports
+from six.moves import range
+from six.moves import zip
+
+
+def uniform_positive_integers_with_sum(count, sum_):
+  """Returns list of size `count` of integers >= 1, summing to `sum_`."""
+  assert sum_ >= 0
+  if count > sum_:
+    raise ValueError('Cannot find {} numbers >= 1 with sum {}'
+                     .format(count, sum_))
+  if count == 0:
+    return []
+  # Select `count - 1` numbers from {1, ..., sum_ - 1}
+  separators = random.sample(list(range(1, sum_)), count - 1)
+  separators = sorted(separators)
+  return [right - left
+          for left, right in zip([0] + separators, separators + [sum_])]
+
+
+def uniform_non_negative_integers_with_sum(count, sum_):
+  """Returns list of size `count` of integers >= 0, summing to `sum_`."""
+  positive = uniform_positive_integers_with_sum(count, sum_ + count)
+  return [i - 1 for i in positive]
+
+
+def log_number_binary_trees(size):
+  """Returns (nat) log of number of binary trees with `size` internal nodes."""
+  # This is equal to log of C_size, where C_n is the nth Catalan number.
+  assert isinstance(size, int)
+  assert size >= 0
+  log = 0.0
+  for k in range(2, size + 1):
+    log += math.log(size + k) - math.log(k)
+  return log
diff --git a/mathematics_dataset/mathematics_dataset/util/combinatorics_test.py b/mathematics_dataset/mathematics_dataset/util/combinatorics_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..def98972e63b9e50c5c8cd7d07d7520b3c642046
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/combinatorics_test.py
@@ -0,0 +1,64 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests mathematics_dataset.util.combinatorics."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+
+# Dependency imports
+from absl.testing import absltest
+from util import combinatorics
+
+
+class CombinatoricsTest(absltest.TestCase):
+
+  def testPositiveIntegersWithSum(self):
+    result = combinatorics.uniform_positive_integers_with_sum(1, 1)
+    self.assertEqual(result, [1])
+    result = combinatorics.uniform_positive_integers_with_sum(2, 2)
+    self.assertEqual(result, [1, 1])
+    result = combinatorics.uniform_positive_integers_with_sum(1, 10)
+    self.assertEqual(sum(result), 10)
+    result = combinatorics.uniform_positive_integers_with_sum(2, 10)
+    self.assertEqual(sum(result), 10)
+    result = combinatorics.uniform_positive_integers_with_sum(0, 0)
+    self.assertEqual(result, [])
+
+  def testNonNegativeIntegersWithSum(self):
+    result = combinatorics.uniform_non_negative_integers_with_sum(1, 0)
+    self.assertEqual(result, [0])
+    result = combinatorics.uniform_non_negative_integers_with_sum(2, 0)
+    self.assertEqual(result, [0, 0])
+    result = combinatorics.uniform_non_negative_integers_with_sum(3, 10)
+    self.assertEqual(sum(result), 10)
+
+  def testLogNumberBinaryTrees(self):
+    self.assertAlmostEqual(
+        combinatorics.log_number_binary_trees(0), math.log(1))
+    self.assertAlmostEqual(
+        combinatorics.log_number_binary_trees(1), math.log(1))
+    self.assertAlmostEqual(
+        combinatorics.log_number_binary_trees(2), math.log(2))
+    self.assertAlmostEqual(
+        combinatorics.log_number_binary_trees(3), math.log(5))
+    self.assertAlmostEqual(
+        combinatorics.log_number_binary_trees(4), math.log(14))
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/util/composition.py b/mathematics_dataset/mathematics_dataset/util/composition.py
new file mode 100644
index 0000000000000000000000000000000000000000..27108d3f4235ee70ab41199b7242bb8fb927603f
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/composition.py
@@ -0,0 +1,596 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""For performing module composition."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import collections
+import random
+import string
+
+# Dependency imports
+from sample import number
+from sample import ops
+from sample import polynomials
+from util import combinatorics
+from util import display
+import numpy as np
+import six
+from six.moves import range
+from six.moves import zip
+import sympy
+
+
+# Allowed symbols. Don't allow "e", as sympy can hang when printing it as a
+# function symbol (and it's reserved for exponent).
+_ALLOWED_SYMBOLS = set(string.ascii_lowercase).difference(set(['e']))
+
+
+class Polynomial(collections.namedtuple('Polynomial', ('coefficients'))):
+  """Value wrapper for a polynomial function.
+
+  Attributes:
+    coefficients: Numpy array of coefficients; see `polynomials.py`.
+  """
+
+  def __new__(cls, coefficients):
+    coefficients = np.asarray(coefficients)
+    return super(Polynomial, cls).__new__(cls, coefficients)
+
+
+def is_polynomial(value):
+  return isinstance(value, Polynomial)
+
+
+def is_integer_polynomial(value):
+  if not is_polynomial(value):
+    return False
+  coefficients = np.reshape(value.coefficients, [-1])
+  return all(number.is_integer(coeff) for coeff in coefficients)
+
+
+# List of pairs of `(filter, sampler)`, where `filter` is a function returning
+# True if the sampler is valid for the given value, and `sampler` returns an
+# `Entity`.
+_FILTERS_AND_SAMPLERS = []
+
+
+def module(filter_):
+  """Returns a Decorator for a module function.
+
+  The returned decorator adds the function to the library of known modules.
+
+  Args:
+    filter_: Callable determining whether the module can handle a given value.
+
+  Returns:
+    Decorator that adds the module function to the library.
+  """
+  def decorator(module_fn):
+    """Decorates a module function."""
+    _FILTERS_AND_SAMPLERS.append((filter_, module_fn))
+    return module_fn
+  return decorator
+
+
+class SampleArgs(
+    collections.namedtuple('SampleArgs', ('num_modules', 'entropy'))):
+  """For sampling mathematical entities / questions."""
+
+  def peel(self, frac=1):
+    """Peels one (or `frac`) of a module's entropy.
+
+    In addition to a portion of the entropy, this returns a new `SampleArgs`
+    (since this object is immutable), which you should use when creating child
+    modules.
+
+    Args:
+      frac: Float; proportion of module's entropy to take.
+
+    Returns:
+      Triple `(entropy, new_sample_args)`, where `new_sample_args` is a new
+      `SampleArgs` with the entropy removed.
+    """
+    entropy = frac * self.entropy / self.num_modules
+    new_sample_args = SampleArgs(
+        num_modules=self.num_modules, entropy=self.entropy - entropy)
+    return entropy, new_sample_args
+
+  def split(self, count):
+    """Splits the entropy and module counts up.
+
+    Args:
+      count: Integer >= 1; the split size.
+
+    Returns:
+      List of `SampleArgs` of length `count`, to be passed to create child
+      entities.
+
+    Raises:
+      ValueError: If it was not possible to use up all the entropy, for example,
+          all requested types were `WithValue`.
+    """
+    num_child_modules = self.num_modules - 1
+
+    # Sample module counts at random
+    module_counts = combinatorics.uniform_non_negative_integers_with_sum(
+        count, num_child_modules)
+
+    if num_child_modules == 0:
+      if self.entropy > 0:
+        raise ValueError('Unused entropy')
+      entropies = np.zeros(count)
+    else:
+      entropies = self.entropy * np.random.dirichlet(
+          np.maximum(1e-9, module_counts))
+
+    sample_args = []
+    for i, num_modules in enumerate(module_counts):
+      child_sample_args = SampleArgs(
+          num_modules=num_modules, entropy=entropies[i])
+      sample_args.append(child_sample_args)
+
+    return sample_args
+
+
+class PreSampleArgs(
+    collections.namedtuple(
+        'PreSampleArgs',
+        ('min_modules', 'max_modules', 'min_entropy', 'max_entropy'))):
+  """Sample args before module count and entropy have been sampled."""
+
+  def __call__(self):
+    """Samples `SampleArgs`."""
+    return SampleArgs(
+        num_modules=random.randint(self.min_modules, self.max_modules),
+        entropy=random.uniform(self.min_entropy, self.max_entropy))
+
+  def peel(self, *args, **kwargs):
+    sample_args = self()
+    return sample_args.peel(*args, **kwargs)
+
+  def split(self, *args, **kwargs):
+    sample_args = self()
+    return sample_args.split(*args, **kwargs)
+
+
+class FunctionHandle(object):
+  """Special handle to allow function composition.
+
+  For example, suppose fn1 = f o g
+                       fn2 = h
+
+  and we want to display fn1(fn2(x)) = f(g(h(x))). This function basically just
+  stores the list of sympy functions.
+  """
+
+  def __init__(self, *function_entities):
+    """Initialize a `FunctionHandle`.
+
+    Args:
+      *function_entities: List of function letters and `Entity`s representing
+          functions, to be composed.
+    """
+    self._functions = []
+    for fn in function_entities:
+      if isinstance(fn, str):
+        functions = [sympy.Function(fn)]
+      else:
+        assert isinstance(fn, Entity)
+        assert isinstance(fn.handle, FunctionHandle)
+        functions = fn.handle.functions
+      self._functions += functions
+
+  def apply(self, *input_):
+    """Returns f(g(...(input)...)) where f, g, ... are the functions."""
+    result = None
+    for function in reversed(self._functions):
+      if result is None:
+        result = function(*input_)
+      else:
+        result = function(result)
+    return result
+
+  @property
+  def functions(self):
+    return self._functions
+
+  def __str__(self):
+    raise ValueError('This should not be directly converted to a string')
+
+
+def _polynomial_entity(value, context):
+  """Create a generic `Entity` describing a polynomial."""
+  assert isinstance(value, Polynomial)
+  coefficients = np.asarray(value.coefficients)
+  num_variables = coefficients.ndim
+  variables = [sympy.Symbol(context.pop()) for _ in range(num_variables)]
+  function_symbol = context.pop()
+  handle = FunctionHandle(function_symbol)
+  handle_description = sympy.Function(function_symbol)(*variables)
+
+  polynomial = polynomials.coefficients_to_polynomial(coefficients, variables)
+  polynomial = polynomial.sympy()
+
+  return Entity(
+      context=context,
+      value=value,
+      expression=polynomial,
+      polynomial_variables=variables,
+      description='Seja {function} = {polynomial}.',
+      handle=handle,
+      function=handle_description,
+      polynomial=polynomial)
+
+
+class Context(object):
+  """Keeps track of used symbols, and sampling of children.
+
+  Each context is associated with an entity. Entities are constructed in a
+  tree-like fashion.
+  """
+
+  def __init__(self, relation_symbols=None):
+    """Initializes a `Context`.
+
+    Args:
+      relation_symbols: Set of symbols used "externally": this means (with
+          reference to the tree in which this context sits) all symbols
+          occurring that aren't in this node or sub-nodes.
+    """
+    if relation_symbols is None:
+      relation_symbols = set()
+    else:
+      assert isinstance(relation_symbols, set)
+    for symbol in relation_symbols:
+      assert isinstance(symbol, str)
+
+    self._relation_symbols = relation_symbols
+    self._self_symbols = set()
+    self._child_symbols = set()
+    self._module_count = 1
+    self._child_entities = []
+
+  @property
+  def relation_symbols(self):
+    return self._relation_symbols.copy()
+
+  @property
+  def self_symbols(self):
+    return self._self_symbols.copy()
+
+  @property
+  def child_symbols(self):
+    return self._child_symbols.copy()
+
+  @property
+  def child_entities(self):
+    return self._child_entities[:]
+
+  def pop(self):
+    """Returns an unused symbol (and keeps track of it being used)."""
+    allowed = (_ALLOWED_SYMBOLS
+               .difference(self._relation_symbols)
+               .difference(self._self_symbols)
+               .difference(self._child_symbols))
+    if not allowed:
+      raise ValueError('Ran out of symbols')
+    symbol = random.choice(list(allowed))
+    self._self_symbols.add(symbol)
+    return symbol
+
+  def mark_used(self, symbol):
+    """Marks a given symbol as used."""
+    assert isinstance(symbol, str)
+    if (symbol in self._relation_symbols
+        or symbol in self._self_symbols
+        or symbol in self._child_symbols):
+      raise ValueError('Symbol {} already used'.format(symbol))
+    self._self_symbols.add(symbol)
+
+  @property
+  def module_count(self):
+    """Returns the number of modules sampled."""
+    return self._module_count
+
+  def _sampler(self, value, sample_args):
+    """Returns a sampler appropriate for the given args.
+
+    Args:
+      value: Target value.
+      sample_args: Instance of `SampleArgs` controlling entropy etc.
+
+    Returns:
+      A sampler for producing entities.
+
+    Raises:
+      ValueError: If no valid samplers were found.
+    """
+    valid = []
+    for filter_, sampler in _FILTERS_AND_SAMPLERS:
+      if filter_(value):
+        valid.append(sampler)
+    if not valid:
+      raise ValueError('No valid samplers found: value={} sample_args={}'
+                       .format(value, sample_args))
+    return random.choice(valid)
+
+  def _value_entity(self, value, context):
+    if isinstance(value, (sympy.Integer, sympy.Rational, display.Decimal)):
+      return Entity(context=context, value=value, handle=value)
+    if isinstance(value, Polynomial):
+      return _polynomial_entity(value, context)
+    raise ValueError('Don\'t know how to handle value={} of type {}'
+                     .format(value, type(value)))
+
+  def sample(self, sample_args, values):
+    """Sample multiple entities.
+
+    Args:
+      sample_args: Instance of `SampleArgs`. The min and max entropy
+          and module count will be split up betwene the various entities
+          sampled.
+      values: List of values to sample.
+
+    Returns:
+      List of `Entity` of the same length as `types`.
+
+    Raises:
+      RuntimeError: If one of the modules generates a non-`Entity`.
+    """
+    # Can only sample children once.
+    assert self._module_count == 1
+    assert not self._child_symbols
+    assert not self._child_entities
+
+    if isinstance(sample_args, PreSampleArgs):
+      sample_args = sample_args()
+    sample_args_split = sample_args.split(len(values))
+
+    def all_symbols():
+      return (self._relation_symbols
+              .union(self._self_symbols)
+              .union(self._child_symbols))
+
+    for value, child_sample_args in zip(values, sample_args_split):
+      if number.is_integer(value):
+        value = sympy.Integer(value)
+
+      all_symbols_ = all_symbols()
+      context = Context(all_symbols_)
+
+      if child_sample_args.num_modules == 0:
+        entity = self._value_entity(value, context)
+      else:
+        sampler = self._sampler(value, child_sample_args)
+        entity = sampler(value, child_sample_args, context)
+        if not isinstance(entity, Entity):
+          raise RuntimeError(
+              'Expected entity, but got {} instead'.format(entity))
+        if (not number.is_integer_or_rational_or_decimal(entity.value)
+            and not isinstance(entity.value, Polynomial)):
+          raise RuntimeError('sampler {} returned invalid value of type {}'
+                             .format(sampler, type(entity.value)))
+        if ((number.is_integer_or_rational_or_decimal(value)
+             and entity.value != value)
+            or (isinstance(value, Polynomial) and not np.array_equal(
+                entity.value.coefficients, value.coefficients))):
+          raise RuntimeError(
+              'entity values differ, sampler={} wanted={} got={}'
+              .format(sampler, value, entity.value))
+        if child_sample_args.num_modules != context.module_count:
+          raise RuntimeError(
+              'unused modules, value={} sample_args={} context.module_count={},'
+              ' sampler={}'
+              .format(value, child_sample_args, context.module_count, sampler))
+        self._module_count += context.module_count
+
+      self._child_entities.append(entity)
+      for symbol in context.self_symbols.union(context.child_symbols):
+        assert symbol not in all_symbols_
+        self._child_symbols.add(symbol)
+
+    return self._child_entities
+
+  def sample_by_replacing_constants(self, sample_args, expressions):
+    """Replaces some of the constants with handles from other modules."""
+    max_children = sample_args.num_modules - 1
+    if max_children <= 0:
+      return
+    if isinstance(expressions, ops.Op):
+      expressions = [expressions]
+    constants = ops.number_constants(expressions)
+    if not constants:
+      raise ValueError('No constants to replace in {}'
+                       .format([str(expr) for expr in expressions]))
+
+    sample_count = random.randint(1, min(max_children, len(constants)))
+    constants = random.sample(constants, sample_count)
+
+    values = [constant.value for constant in constants]
+    entities = self.sample(sample_args, values)
+    for constant, entity in zip(constants, entities):
+      constant.value = entity.handle
+
+
+def expand_entities(context, **kwargs):
+  """Returns prefix description and updated `**kwargs`.
+
+  Args:
+    context: Instance of `Context` containing all `Entity`s that occur.
+    **kwargs: Dictionary of key/value pairs, some of which are `Entity`s.
+
+  Returns:
+    Pair `(child_description, new_kwargs)`. `child_description` is a description
+    of the entities contained in `kwargs`, and `new_kwargs` contains handles.
+  """
+  kwargs = kwargs.copy()
+  entities = set(context.child_entities)
+  for key, maybe_entity in six.iteritems(kwargs):
+    if isinstance(maybe_entity, Entity):
+      entities.add(maybe_entity)
+      kwargs[key] = maybe_entity.handle
+  entities = list(entities)
+  random.shuffle(entities)
+
+  child_descriptions = []
+  for entity in entities:
+    child_descriptions.append(entity.child_description)
+    if not entity.expression_used:
+      child_descriptions.append(entity.description)
+
+  child_description = ' '.join([s for s in child_descriptions if s])
+  return child_description, kwargs
+
+
+class Entity(object):
+  """An entity (e.g., representing an integer or function).
+
+  Example usage:
+
+  ```
+  new_entity = Entity(
+      context=context,
+      value=17,
+      description='Let {self} be the gcd of {q} and {q}.'
+      p=p, q=q)
+  ```
+
+  Above, {self} will be replaced with a new symbol, to be used as the entity
+  handle.
+  """
+
+  def __init__(self, context, value, description='', handle=None,
+               expression=None, polynomial_variables=None,
+               **description_kwargs):
+    """Initializes an entity.
+
+    Args:
+      context: Instance of `Context` keeping track of used symbols and child
+          entities (that may not explicitly occur in `description_kwargs`).
+      value: The value represented by this Entity.
+      description: String, describing this Entity. This can contain '{self}' as
+          a substring, in which case it will be substituted with a new handle
+          (and `context` must be non-None).
+      handle: Optional string/symbol that refers to this Entity. This must
+          always be provided if '{self}' does not occur in `description`.
+      expression: Optional string (or something that can be converted to a
+          string, like ops.Op or sympy.expr) representing the value of this
+          Entity, that can be used directly instead of using `handle`.
+      polynomial_variables: If `expression` provided, then for polynomial
+          entities, the variables used.
+      **description_kwargs: Dict of substitutions (including entities) for use
+          in `description`.
+
+    Raises:
+      ValueError: If handle specified and '{self}' in description; or neither.
+    """
+    self._value = value
+
+    child_description, description_kwargs = expand_entities(
+        context, **description_kwargs)
+
+    if '{self}' in description:
+      if handle is not None:
+        raise ValueError('Cannot specify handle if {self} in description')
+      handle = context.pop()
+      description_kwargs['self'] = handle
+      handle = sympy.var(handle)
+    else:
+      if handle is None:
+        raise ValueError('Must specify handle if {self} not in description')
+      if isinstance(handle, str):
+        handle = sympy.var(handle)
+
+    if (isinstance(value, Polynomial)
+        and expression is not None
+        and polynomial_variables is None):
+      raise ValueError('Must provided polynomial_variables')
+
+    self._child_description = child_description
+    self._description = description.format(**description_kwargs)
+    self._handle = handle
+    self._expression = expression
+    self._polynomial_variables = polynomial_variables
+
+    # For checking that we don't use both the handle and the expression form of
+    # this entity.
+    self._handle_used = False
+    self._expression_used = False
+
+  @property
+  def value(self):
+    """The actual value of the entity (e.g., sympy object)."""
+    return self._value
+
+  @property
+  def child_description(self):
+    """A string describing the entities that this Entity relies on."""
+    return self._child_description
+
+  @property
+  def description(self):
+    """A string describing the entity."""
+    assert not self._expression_used
+    self._handle_used = True
+    return self._description
+
+  def has_expression(self):
+    """Returns whether there is an expression to use (instead of handle)."""
+    return self._expression is not None
+
+  @property
+  def handle(self):
+    """A "handle", e.g., SymPy symbol or `function.CompositionHandle`."""
+    assert not self._expression_used
+    self._handle_used = True
+    return self._handle
+
+  @property
+  def expression(self):
+    """An expression representing the entity; possibly None.
+
+    If this is used, then `description` does not need to get used.
+
+    Returns:
+      String or None.
+    """
+    assert not self._handle_used
+    self._expression_used = True
+    return self._expression
+
+  @property
+  def polynomial_variables(self):
+    """For when `expression` is not None, and this entity is a polynomial."""
+    return self._polynomial_variables
+
+  @property
+  def expression_used(self):
+    """Returns true if expression instead of handle was used."""
+    return self._expression_used
+
+  @property
+  def expression_else_handle(self):
+    """Returns an expression if present, else the handle."""
+    if self.has_expression():
+      return self.expression
+    else:
+      return self.handle
+
+  def __str__(self):
+    """Raises value error - should not attempt to convert to string directly."""
+    raise ValueError('Should not convert Entity directly to string')
diff --git a/mathematics_dataset/mathematics_dataset/util/composition_test.py b/mathematics_dataset/mathematics_dataset/util/composition_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..0d3e533bb8d718a5805600611a36cd2043e36599
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/composition_test.py
@@ -0,0 +1,65 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.util.composition."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from util import composition
+import sympy
+
+
+class FunctionHandleTest(absltest.TestCase):
+
+  def testApply(self):
+    handle = composition.FunctionHandle('f', 'g')
+    applied = handle.apply(*sympy.symbols('x y'))
+    self.assertEqual(str(applied), 'f(g(x, y))')
+    applied = handle.apply(sympy.symbols('x'))
+    self.assertEqual(str(applied), 'f(g(x))')
+
+
+class ContextTest(absltest.TestCase):
+
+  def testPeel(self):
+    sample_args = composition.SampleArgs(4, 3.0)
+    entropy, new_sample_args = sample_args.peel()
+    self.assertAlmostEqual(entropy, 0.75)
+    self.assertEqual(new_sample_args.num_modules, 4)
+    self.assertAlmostEqual(new_sample_args.entropy, 2.25)
+
+  def testSplit(self):
+    sample_args = composition.SampleArgs(4, 5.0)
+    children = sample_args.split(2)
+    self.assertLen(children, 2)
+    self.assertEqual(sum([child.num_modules for child in children]), 3)
+    self.assertAlmostEqual(sum([child.entropy for child in children]), 5.0)
+
+
+class EntityTest(absltest.TestCase):
+
+  def testInit_valueErrorIfSelfAndHandle(self):
+    with self.assertRaisesRegex(self, ValueError, 'Cannot specify handle'):
+      composition.Entity(context=composition.Context(),
+                         value=0,
+                         description='Something with {self}. ',
+                         handle='additional')
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/util/display.py b/mathematics_dataset/mathematics_dataset/util/display.py
new file mode 100644
index 0000000000000000000000000000000000000000..0ab5b2e44356913752606a77d86e70824f1030d9
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/display.py
@@ -0,0 +1,404 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Functionality for displaying expressions.
+
+SymPy provides a lot of functionality for displaying expressions, but it's
+slightly too centered on being a symbolic maths engine to provides all our
+needs. For example, it's impossible to display an unsimplified fraction like
+3/6, or a decimal that isn't internally represented as a float and thus subject
+to rounding.
+
+Also provides some other convenience such as converting numbers to words, and
+displaying percentages (properly formatted).
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import decimal
+
+# Dependency imports
+import sympy
+
+
+# For converting integers to words:
+_INTEGER_LOW = [
+    'zero', 'um', 'dois', 'três', 'quatro', 'cinco', 'seis', 'sete', 'oito',
+    'nove', 'dez', 'onze', 'doze', 'treze', 'quatorze', 'quinze',
+    'dezasseis', 'dezassete', 'dezoito', 'dezanove'
+]
+_INTEGER_MID = [
+    '', '', 'vinte', 'trinta', 'quarenta', 'cinquenta', 'sessenta', 'setenta', 'oitenta',
+    'noventa'
+]
+_INTEGER_HIGH = [
+    (int(1e12), 'bilião'), (int(1e9), 'milhar de milhão'), (int(1e6), 'milhão'),
+    (int(1e3), 'milhar'), (100, 'centena')
+]
+
+
+# For converting rationals to words:
+_SINGULAR_DENOMINATORS = [
+    '', '', 'metade', 'terço', 'quarto', 'quinto', 'sexto', 'sétimo', 'oitavo',
+    'nono', 'dez avos', 'onze avos', 'onze avos', 'treze avos', 'quatorze avos',
+    'quinze avos', 'dezasseis avos', 'dezassete avos', 'dezoito avos', 'dezanove avos',
+    'vinte avos'
+]
+_PLURAL_DENOMINATORS = [
+    '', '', 'metades', 'terços', 'quartos', 'quintos', 'sextos', 'sétimos',
+    'oitavos', 'nonos', 'décimos', 'onze avos', 'doze avos', 'treze avos',
+    'quatorze avos', 'quinze avos', 'dezasseis avos', 'dezassete avos', 'dezoito avos',
+    'dezanove avos', 'vinte avos'
+]
+
+
+# For converting ordinals to words:
+_ORDINALS = [
+    'zerésima', 'primeira', 'segunda', 'terceira', 'quarta', 'quinta', 'sexta', 'sétima',
+    'oitava', 'nona', 'décima', 'décima primeira', 'décima segunda', 'décima terceira',
+    'décima quarta', 'décima quinta', 'décima sexta', 'décima sétima', 'décima oitava',
+    'décima nona', 'vigésima'
+]
+
+
+class Decimal(object):
+  """Display a value as a decimal."""
+
+  def __init__(self, value):
+    """Initializes a `Decimal`.
+
+    Args:
+      value: (Sympy) value to display as a decimal.
+
+    Raises:
+      ValueError: If `value` cannot be represented as a non-terminating decimal.
+    """
+    self._value = sympy.Rational(value)
+
+    numer = int(sympy.numer(self._value))
+    denom = int(sympy.denom(self._value))
+
+    denom_factors = list(sympy.factorint(denom).keys())
+    for factor in denom_factors:
+      if factor not in [2, 5]:
+        raise ValueError('Cannot represent {} as a non-recurring decimal.'
+                         .format(value))
+    self._decimal = decimal.Decimal(numer) / decimal.Decimal(denom)
+
+  @property
+  def value(self):
+    """Returns the value as a `sympy.Rational` object."""
+    return self._value
+
+  def _sympy_(self):
+    return self._value
+
+  def decimal_places(self):
+    """Returns the number of decimal places, e.g., 32 has 0 and 1.43 has 2."""
+    if isinstance(self._decimal, int):
+      return 0
+    elif isinstance(self._decimal, decimal.Decimal):
+      return -self._decimal.as_tuple().exponent
+
+  def __str__(self):
+    sign, digits, exponent = self._decimal.as_tuple()
+    sign = '' if sign == 0 else '-'
+
+    num_left_digits = len(digits) + exponent  # number digits "before" point
+
+    if num_left_digits > 0:
+      int_part = ''.join(str(digit) for digit in digits[:num_left_digits])
+    else:
+      int_part = '0'
+
+    if exponent < 0:
+      frac_part = '.'
+      if num_left_digits < 0:
+        frac_part += '0' * -num_left_digits
+      frac_part += ''.join(str(digit) for digit in digits[exponent:])
+    else:
+      frac_part = ''
+
+    return sign + int_part + frac_part
+
+  def __add__(self, other):
+    if not isinstance(other, Decimal):
+      raise ValueError('Arithmetic support limited to other `Decimal`s.')
+    return Decimal(self.value + other.value)
+
+  def __sub__(self, other):
+    if not isinstance(other, Decimal):
+      raise ValueError('Arithmetic support limited to other `Decimal`s.')
+    return Decimal(self.value - other.value)
+
+  def __mul__(self, other):
+    if not isinstance(other, Decimal):
+      raise ValueError('Arithmetic support limited to other `Decimal`s.')
+    return Decimal(self.value * other.value)
+
+  def __neg__(self):
+    return Decimal(-self.value)
+
+  def round(self, ndigits=0):
+    """Returns a new `Decimal` rounded to this many decimal places."""
+    scale = sympy.Integer(10 ** ndigits)
+    numer = sympy.numer(self.value) * scale
+    denom = sympy.denom(self.value)
+    return Decimal(int(round(numer / denom)) / scale)
+
+  def __round__(self, ndigits):
+    return self.round(ndigits)
+
+  def __int__(self):
+    """Returns conversion to integer if possible; TypeError if non-integer."""
+    if self.decimal_places() == 0:
+      return int(self._decimal)
+    else:
+      raise TypeError('Cannot represent {} as an integer.'.format(str(self)))
+
+  # NOTE: this is implemented in addition to `__cmp__` because SymPy does not
+  # support inequality comparison between sympy objects and objects that are not
+  # convertible to sympy objects (such as strings).
+  def __eq__(self, other):
+    return self.value == other
+
+  # Python 2 comparison
+  def __cmp__(self, other):
+    if self.value == other:
+      return 0
+    if self.value < other:
+      return -1
+    return 1
+
+  # Python 3 comparison:
+  def __lt__(self, other):
+    return self.value < other
+
+  def __le__(self, other):
+    return self.value <= other
+
+  def __gt__(self, other):
+    return self.value > other
+
+  def __ge__(self, other):
+    return self.value >= other
+
+
+class Percentage(object):
+  """Container for a percentage."""
+
+  def __init__(self, value):
+    """Initializes a `Percentage`.
+
+    Args:
+      value: Percentage as a fractional value. E.g., pass in
+          `sympy.Rational(2, 5)` to create the percentage "40%".
+    """
+    self._value = value
+
+  def _sympy_(self):
+    return self._value
+
+  def __str__(self):
+    # Display percentages as decimals (not fractions).
+    value = Decimal(self._value * 100)
+    return str(value) + '%'
+
+
+class NonSimpleRational(object):
+  """Container for rational a / b where allow gcd(a, b) > 1."""
+
+  def __init__(self, numer, denom):
+    self._numer = numer
+    self._denom = denom
+
+  @property
+  def numer(self):
+    return self._numer
+
+  @property
+  def denom(self):
+    return self._denom
+
+  def __str__(self):
+    return '{}/{}'.format(self._numer, self._denom)
+
+
+class StringNumber(object):
+  """A string representing a number, that can also be sympified."""
+
+  def __init__(self, value, join_number_words_with_hyphens=True):
+    """Initializes a `StringNumber`.
+
+    Args:
+      value: An integer or rational.
+      join_number_words_with_hyphens: Whether to join the words in integers with
+          hyphens when describing as a string.
+    """
+    self._join_number_words_with_hyphens = join_number_words_with_hyphens
+    self._sympy_value = sympy.sympify(value)
+    self._string = self._to_string(value)
+
+  def _integer_to_words(self, integer):
+    """Converts an integer to a list of words."""
+    if integer < 0:
+      raise ValueError('Cannot handle negative numbers.')
+
+    if integer < 20:
+      return [_INTEGER_LOW[integer]]
+
+    words = None
+
+    if integer < 100:
+      tens, ones = divmod(integer, 10)
+      if ones > 0:
+        return [_INTEGER_MID[tens], _INTEGER_LOW[ones]]
+      else:
+        return [_INTEGER_MID[tens]]
+
+    for value, word in _INTEGER_HIGH:
+      if integer >= value:
+        den, rem = divmod(integer, value)
+        words = self._integer_to_words(den) + [word]
+        if rem > 0:
+          if rem < 100:
+            words.append('and')
+          words += self._integer_to_words(rem)
+        return words
+
+  def _rational_to_string(self, rational):
+    """Converts a rational to words, e.g., "two thirds"."""
+    numer = sympy.numer(rational)
+    denom = sympy.denom(rational)
+
+    numer_words = self._to_string(numer)
+
+    if denom == 1:
+      return numer_words
+
+    if denom <= 0 or denom >= len(_PLURAL_DENOMINATORS):
+      raise ValueError('Unsupported denominator {}.'.format(denom))
+
+    if numer == 1:
+      denom_word = _SINGULAR_DENOMINATORS[denom]
+    else:
+      denom_word = _PLURAL_DENOMINATORS[denom]
+
+    return '{} {}'.format(numer_words, denom_word)
+
+  def _to_string(self, number):
+    """Converts an integer or rational to words."""
+    if isinstance(number, sympy.Integer) or isinstance(number, int):
+      words = self._integer_to_words(number)
+      join_char = '-' if self._join_number_words_with_hyphens else ' '
+      return join_char.join(words)
+    elif isinstance(number, sympy.Rational):
+      return self._rational_to_string(number)
+    else:
+      raise ValueError('Unable to handle number {} with type {}.'
+                       .format(number, type(number)))
+
+  def _sympy_(self):
+    return self._sympy_value
+
+  def __str__(self):
+    return self._string
+
+
+class StringOrdinal(object):
+  """A string representation of an ordinal, e.g., "first"."""
+
+  def __init__(self, position):
+    """Initializes a `StringOrdinal`.
+
+    Args:
+      position: An integer >= 0.
+
+    Raises:
+      ValueError: If `position` is non-positive or out of range.
+    """
+    if position < 0 or position >= len(_ORDINALS):
+      raise ValueError('Unsupported ordinal {}.'.format(position))
+    self._string = _ORDINALS[position]
+
+  def __str__(self):
+    return self._string
+
+
+class NumberList(object):
+  """Contains a list of numbers, intended for display."""
+
+  def __init__(self, numbers):
+    self._numbers = numbers
+
+  def __str__(self):
+    """Converts the list to a string.
+
+    Returns:
+      Human readable string.
+
+    Raises:
+      ValueError: if any of the strings contain a comma and thus would lead to
+          an ambigious representation.
+    """
+    strings = []
+    for number in self._numbers:
+      string = str(number)
+      if ',' in string:
+        raise ValueError('String representation of the list will be ambigious, '
+                         'since term "{}" contains a comma.'.format(string))
+      strings.append(string)
+    return ', '.join(strings)
+
+
+class NumberInBase(object):
+  """Contains value, represented in a given base."""
+
+  def __init__(self, value, base):
+    """Initializes a `NumberInBase`.
+
+    Args:
+      value: Positive or negative integer.
+      base: Integer in the range [2, 36].
+
+    Raises:
+      ValueError: If base is not in the range [2, 36] (since this is the limit
+          that can be represented by 10 numbers plus 26 letters).
+    """
+    if not 2 <= base <= 36:
+      raise ValueError('base={} must be in the range [2, 36]'.format(base))
+    self._value = value
+    self._base = base
+
+    chars = []
+    remainder = abs(value)
+    while True:
+      digit = remainder % base
+      char = str(digit) if digit <= 9 else chr(ord('a') + digit - 10)
+      chars.append(char)
+      remainder = int(remainder / base)
+      if remainder == 0:
+        break
+    if value < 0:
+      chars.append('-')
+
+    self._str = ''.join(reversed(chars))
+
+  def __str__(self):
+    return self._str
+
+  def _sympy_(self):
+    return self._value
diff --git a/mathematics_dataset/mathematics_dataset/util/display_test.py b/mathematics_dataset/mathematics_dataset/util/display_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..6520655588a61631db6777c40b97a6d0b5d4c44b
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/display_test.py
@@ -0,0 +1,207 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.util.display."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from util import display
+import sympy
+
+
+class DecimalTest(absltest.TestCase):
+
+  def testBasic_integer(self):
+    decimal = display.Decimal(123)
+    self.assertEqual(str(decimal), '123')
+    self.assertEqual(sympy.sympify(decimal), sympy.Integer(123))
+    self.assertEqual(decimal.decimal_places(), 0)
+
+  def testBasic_ten(self):
+    decimal = display.Decimal(10)
+    self.assertEqual(str(decimal), '10')
+    self.assertEqual(sympy.sympify(decimal), sympy.Integer(10))
+    self.assertEqual(decimal.decimal_places(), 0)
+
+  def testBasic(self):
+    decimal = display.Decimal(sympy.Rational(123, 100))
+    self.assertEqual(str(decimal), '1.23')
+    self.assertEqual(sympy.sympify(decimal), sympy.Rational(123, 100))
+    self.assertEqual(decimal.decimal_places(), 2)
+
+  def testStr(self):
+    self.assertEqual(str(display.Decimal(sympy.Rational(0, 10))), '0')
+    self.assertEqual(str(display.Decimal(sympy.Rational(-1, 10))), '-0.1')
+    self.assertEqual(str(display.Decimal(sympy.Rational(-11, 10))), '-1.1')
+    self.assertEqual(str(display.Decimal(sympy.Rational(11, 10))), '1.1')
+    self.assertEqual(str(display.Decimal(sympy.Rational(101, 1))), '101')
+    self.assertEqual(
+        str(display.Decimal(sympy.Rational(20171, 1000000))), '0.020171')
+
+  def testStr_verySmall(self):
+    # Tests it doesn't display in "scientific" notation 1E-9.
+    decimal = display.Decimal(sympy.Rational(1, 1000000000))
+    self.assertEqual(str(decimal), '0.000000001')
+
+  def testAdd(self):
+    self.assertEqual((display.Decimal(2) + display.Decimal(3)).value, 5)
+
+  def testSub(self):
+    self.assertEqual((display.Decimal(2) - display.Decimal(3)).value, -1)
+
+  def testMul(self):
+    self.assertEqual((display.Decimal(2) * display.Decimal(3)).value, 6)
+
+  def testRound(self):
+    decimal = display.Decimal(sympy.Rational(2675, 1000))  # 2.675
+    self.assertEqual(sympy.sympify(decimal.round()), sympy.Integer(3))
+    self.assertEqual(sympy.sympify(decimal.round(1)), sympy.Rational(27, 10))
+    self.assertEqual(sympy.sympify(decimal.round(2)), sympy.Rational(268, 100))
+    self.assertEqual(sympy.sympify(decimal.round(3)),
+                     sympy.Rational(2675, 1000))
+
+  def testInt(self):
+    decimal = display.Decimal(123)
+    self.assertEqual(int(decimal), 123)
+
+  def testInt_errorIfNonInt(self):
+    decimal = display.Decimal(sympy.Rational(1, 2))
+    with self.assertRaisesRegex(self, TypeError, 'Cannot represent'):
+      int(decimal)
+
+  def testComparison(self):
+    decimal = display.Decimal(sympy.Rational(-1, 2))
+    # pylint: disable=g-generic-assert
+    self.assertFalse(decimal != -0.5)
+    self.assertTrue(decimal != 0)
+    self.assertFalse(decimal < -0.5)
+    self.assertTrue(decimal < 0)
+    self.assertTrue(decimal <= -0.5)
+    self.assertTrue(decimal <= 0)
+    self.assertFalse(decimal > -0.5)
+    self.assertTrue(decimal > -1)
+    self.assertTrue(decimal >= -0.5)
+    self.assertFalse(decimal >= 0)
+    self.assertFalse(decimal == 0)
+    self.assertTrue(decimal == -0.5)
+
+  def testNegation(self):
+    decimal = display.Decimal(sympy.Rational(1, 2))
+    decimal = -decimal
+    self.assertNotEqual(decimal, 0.5)
+    self.assertEqual(decimal, -0.5)
+
+
+class PercentageTest(absltest.TestCase):
+
+  def testPercentage(self):
+    percentage = display.Percentage(1.5)
+    self.assertEqual(str(percentage), '150%')
+
+    percentage = display.Percentage(sympy.Rational(67, 100))
+    self.assertEqual(str(percentage), '67%')
+
+    percentage = display.Percentage(sympy.Rational(67, 1000))
+    self.assertEqual(str(percentage), '6.7%')
+
+
+class NonSimpleRationalTest(absltest.TestCase):
+
+  def testBasic(self):
+    frac = display.NonSimpleRational(4, 6)
+    self.assertEqual(frac.numer, 4)
+    self.assertEqual(frac.denom, 6)
+    self.assertEqual(str(frac), '4/6')
+
+
+class StringNumberTest(absltest.TestCase):
+
+  def testIntegerToWords(self):
+    words = display.StringNumber(0)
+    self.assertEqual(str(words), 'zero')
+    self.assertEqual(sympy.sympify(words), 0)
+
+    words = display.StringNumber(8)
+    self.assertEqual(str(words), 'eight')
+    self.assertEqual(sympy.sympify(words), 8)
+
+    words = display.StringNumber(12)
+    self.assertEqual(str(words), 'twelve')
+    self.assertEqual(sympy.sympify(words), 12)
+
+    words = display.StringNumber(30)
+    self.assertEqual(str(words), 'thirty')
+    self.assertEqual(sympy.sympify(words), 30)
+
+    words = display.StringNumber(100)
+    self.assertEqual(str(words), 'one-hundred')
+    self.assertEqual(sympy.sympify(words), 100)
+
+    words = display.StringNumber(103)
+    self.assertEqual(str(words), 'one-hundred-and-three')
+    self.assertEqual(sympy.sympify(words), 103)
+
+    words = display.StringNumber(15439822)
+    self.assertEqual(str(words), 'fifteen-million-four-hundred-and-thirty-nine'
+                                 '-thousand-eight-hundred-and-twenty-two')
+    self.assertEqual(sympy.sympify(words), 15439822)
+
+  def testRationalToWords(self):
+    words = display.StringNumber(sympy.Rational(2, 3))
+    self.assertEqual(str(words), 'two thirds')
+
+
+class StringOrdinalTest(absltest.TestCase):
+
+  def testBasic(self):
+    ordinal = display.StringOrdinal(0)
+    self.assertEqual(str(ordinal), 'zeroth')
+    ordinal = display.StringOrdinal(10)
+    self.assertEqual(str(ordinal), 'tenth')
+
+  def testCreate_errorIfNegative(self):
+    with self.assertRaisesRegex(self, ValueError, 'Unsupported ordinal'):
+      display.StringOrdinal(-1)
+
+
+class NumberListTest(absltest.TestCase):
+
+  def testBasic(self):
+    numbers = [2, 3, 1]
+    number_list = display.NumberList(numbers)
+    string = str(number_list)
+    self.assertEqual(string, '2, 3, 1')
+
+
+class NumberInBaseTest(absltest.TestCase):
+
+  def testBasic(self):
+    self.assertEqual(str(display.NumberInBase(1, 10)), '1')
+    self.assertEqual(str(display.NumberInBase(-1, 10)), '-1')
+    self.assertEqual(str(display.NumberInBase(1, 2)), '1')
+    self.assertEqual(str(display.NumberInBase(-1, 2)), '-1')
+    self.assertEqual(str(display.NumberInBase(2, 2)), '10')
+    self.assertEqual(str(display.NumberInBase(-2, 2)), '-10')
+    self.assertEqual(str(display.NumberInBase(10, 16)), 'a')
+    self.assertEqual(str(display.NumberInBase(16, 16)), '10')
+    self.assertEqual(str(display.NumberInBase(256, 16)), '100')
+    self.assertEqual(str(display.NumberInBase(-75483, 10)), '-75483')
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/mathematics_dataset/util/probability.py b/mathematics_dataset/mathematics_dataset/util/probability.py
new file mode 100644
index 0000000000000000000000000000000000000000..3016d0523306bebdd9671a104be8c6da51a61170
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/probability.py
@@ -0,0 +1,405 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Functionality for working with probability spaces and random variables.
+
+Basic recap of probability theory, and thus of classes in this file:
+
+*   A probability space is a (finite or infinite) set Omega with a probability
+    measure defined on this.
+*   A random variable is a mapping from a probability space to another measure
+    space.
+*   An event is a measurable set in a sample space.
+
+For example, suppose a bag contains 3 balls: two red balls, and one white ball.
+This could be represented by a discrete probability space of size 3 with
+elements {1, 2, 3}, with equal measure assigned to all 3 elements; and a random
+variable that maps 1->red, 2->red, and 3->white. Then the probability of drawing
+a red ball is the measure in the probability space of the inverse under the
+random variable mapping of {red}, i.e., of {1, 2}, which is 2/3.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import abc
+import itertools
+
+# Dependency imports
+import six
+from six.moves import zip
+import sympy
+
+
+@six.add_metaclass(abc.ABCMeta)
+class Event(object):
+  """Represents an event in a measure space."""
+
+
+@six.add_metaclass(abc.ABCMeta)
+class ProbabilitySpace(object):
+  """Represents a probability space."""
+
+  @abc.abstractmethod
+  def probability(self, event):
+    """Returns the probability of an event."""
+
+
+@six.add_metaclass(abc.ABCMeta)
+class RandomVariable(object):
+  """Random variable; a mapping from a probability space to a measure space."""
+
+  @abc.abstractmethod
+  def __call__(self, event):
+    """Maps an `_Event` in the probability space to one in the sample space."""
+
+  @abc.abstractmethod
+  def inverse(self, event):
+    """Maps event in the sample space back to the inverse in the prob. space."""
+
+
+class DiscreteEvent(Event):
+  """Set of discrete values."""
+
+  def __init__(self, values):
+    self._values = values
+
+  @property
+  def values(self):
+    return self._values
+
+
+class FiniteProductEvent(Event):
+  """Event consisting of cartesian product of events."""
+
+  def __init__(self, events):
+    """Initializes a `FiniteProductEvent`.
+
+    Args:
+      events: Tuple of `Event`s; resulting event will be cartesian product of
+          these.
+    """
+    self._events = events
+
+  @property
+  def events(self):
+    return self._events
+
+  def all_sequences(self):
+    """Returns iterator of sequences by selecting a single event in each coord.
+
+    This assumes that every component event is an instance of `DiscreteEvent`.
+
+    Returns:
+      Iterator over tuples of values.
+
+    Raises:
+      ValueError: If one of the component events is not a `DiscreteEvent`.
+    """
+    if not all(isinstance(event, DiscreteEvent) for event in self._events):
+      raise ValueError('Not all component events are DiscreteEvents')
+    values_list = [event.values for event in self._events]
+    return itertools.product(*values_list)
+
+
+class CountLevelSetEvent(Event):
+  """Event of all sequences with fixed number of different values occurring."""
+
+  def __init__(self, counts):
+    """Initializes `CountLevelSetEvent`.
+
+    E.g., to construct the event of getting two red balls and one green ball,
+    pass `counts = {red: 2, green: 1}`. (Then `all_sequences()` would return
+    `[(red, red, green), (red, green, red), (green, red, red)]`.
+
+    Args:
+      counts: Dictionary mapping values to the number of times they occur in a
+          sequence.
+    """
+    self._counts = counts
+    self._all_sequences = None
+
+  @property
+  def counts(self):
+    return self._counts
+
+  def all_sequences(self):
+    """Returns all sequences generated by this level set."""
+    if self._all_sequences is None:
+      # Generate via dynamic programming.
+      cache = {}  # dict mapping tuple -> list of tuples
+      labels = list(self._counts.keys())
+
+      def generate(counts):
+        """Returns list of tuples for given `counts` of labels."""
+        if sum(counts) == 0:
+          return [()]
+        counts = tuple(counts)
+        if counts in cache:
+          return cache[counts]
+        generated = []
+        for i, count in enumerate(counts):
+          if count == 0:
+            continue
+          counts_minus = list(counts)
+          counts_minus[i] -= 1
+          counts_minus = tuple(counts_minus)
+          extensions = generate(counts_minus)
+          generated += [tuple([labels[i]] + list(extension))
+                        for extension in extensions]
+        cache[counts] = generated
+        return generated
+
+      self._all_sequences = generate(list(self._counts.values()))
+
+    return self._all_sequences
+
+
+class SequenceEvent(Event):
+  """Collection of sequences."""
+
+  def __init__(self, sequences):
+    self._sequences = sequences
+
+  def all_sequences(self):
+    return self._sequences
+
+
+def normalize_weights(weights):
+  """Normalizes the weights (as sympy.Rational) in dictionary of weights."""
+  weight_sum = sum(six.itervalues(weights))
+  return {
+      i: sympy.Rational(weight, weight_sum)
+      for i, weight in six.iteritems(weights)
+  }
+
+
+class DiscreteProbabilitySpace(ProbabilitySpace):
+  """Discrete probability space."""
+
+  def __init__(self, weights=None):
+    """Initializes an `DiscreteProbabilitySpace`.
+
+    Args:
+      weights: Dictionary mapping values to relative probability of selecting
+          that value. This will be normalized.
+    """
+    self._weights = normalize_weights(weights)
+
+  def probability(self, event):
+    if isinstance(event, DiscreteEvent):
+      return sum(self._weights[value]
+                 for value in event.values if value in self._weights)
+    else:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+  @property
+  def weights(self):
+    """Returns dictionary of probability of each element."""
+    return self._weights
+
+
+class FiniteProductSpace(ProbabilitySpace):
+  """Finite cartesian product of probability spaces."""
+
+  def __init__(self, spaces):
+    """Initializes a `FiniteProductSpace`.
+
+    Args:
+      spaces: List of `ProbabilitySpace`.
+    """
+    self._spaces = spaces
+
+  def all_spaces_equal(self):
+    return all([self._spaces[0] == space for space in self._spaces])
+
+  def probability(self, event):
+    # Specializations for optimization.
+    if isinstance(event, FiniteProductEvent):
+      assert len(self._spaces) == len(event.events)
+      return sympy.prod([
+          space.probability(event_slice)
+          for space, event_slice in zip(self._spaces, event.events)])
+
+    if isinstance(event, CountLevelSetEvent) and self.all_spaces_equal():
+      space = self._spaces[0]
+      counts = event.counts
+      probabilities = {
+          value: space.probability(DiscreteEvent({value}))
+          for value in six.iterkeys(counts)
+      }
+
+      num_events = sum(six.itervalues(counts))
+      assert num_events == len(self._spaces)
+      # Multinomial coefficient:
+      coeff = (
+          sympy.factorial(num_events) / sympy.prod(
+              [sympy.factorial(i) for i in six.itervalues(counts)]))
+      return coeff * sympy.prod([
+          pow(probabilities[value], counts[value])
+          for value in six.iterkeys(counts)
+      ])
+
+    raise ValueError('Unhandled event type {}'.format(type(event)))
+
+  @property
+  def spaces(self):
+    """Returns list of spaces."""
+    return self._spaces
+
+
+class SampleWithoutReplacementSpace(ProbabilitySpace):
+  """Probability space formed by sampling discrete space without replacement."""
+
+  def __init__(self, weights, n_samples):
+    """Initializes a `SampleWithoutReplacementSpace`.
+
+    Args:
+      weights: Dictionary mapping values to relative probability of selecting
+          that value. This will be normalized.
+      n_samples: Number of samples to draw.
+
+    Raises:
+      ValueError: If `n_samples > len(weights)`.
+    """
+    if n_samples > len(weights):
+      raise ValueError('n_samples is more than number of discrete elements')
+    self._weights = normalize_weights(weights)
+    self._n_samples = n_samples
+
+  @property
+  def n_samples(self):
+    """Number of samples to draw."""
+    return self._n_samples
+
+  def probability(self, event):
+    try:
+      all_sequences = event.all_sequences()
+    except AttributeError:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+    probability_sum = 0
+    for sequence in all_sequences:
+      if len(sequence) != len(set(sequence)):
+        continue  # not all unique, so not "without replacement".
+      p_sequence = 1
+      removed_prob = 0
+      for i in sequence:
+        p = self._weights[i] if i in self._weights else 0
+        if p == 0:
+          p_sequence = 0
+          break
+        p_sequence *= p / (1 - removed_prob)
+        removed_prob += p
+      probability_sum += p_sequence
+    return probability_sum
+
+
+class IdentityRandomVariable(RandomVariable):
+  """Identity map of a probability space."""
+
+  def __call__(self, event):
+    return event
+
+  def inverse(self, event):
+    return event
+
+
+class DiscreteRandomVariable(RandomVariable):
+  """Specialization to discrete random variable.
+
+  This is simply a mapping from a discrete space to a discrete space (dictionary
+  lookup).
+  """
+
+  def __init__(self, mapping):
+    """Initializes `DiscreteRandomVariable` from `mapping` dict."""
+    self._mapping = mapping
+    self._inverse = {}
+    for key, value in six.iteritems(mapping):
+      if value in self._inverse:
+        self._inverse[value].add(key)
+      else:
+        self._inverse[value] = set([key])
+
+  def __call__(self, event):
+    if isinstance(event, DiscreteEvent):
+      return DiscreteEvent({self._mapping[value] for value in event.values})
+    else:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+  def inverse(self, event):
+    if isinstance(event, DiscreteEvent):
+      set_ = set()
+      for value in event.values:
+        if value in self._inverse:
+          set_.update(self._inverse[value])
+      return DiscreteEvent(set_)
+    else:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+
+class FiniteProductRandomVariable(RandomVariable):
+  """Product random variable.
+
+  This has the following semantics. Let this be X = (X_1, ..., X_n). Then
+
+  X(w) = (X_1(w_1), ..., X_n(w_n))
+
+  (the sample space is assumed to be of sequence type).
+  """
+
+  def __init__(self, random_variables):
+    """Initializes a `FiniteProductRandomVariable`.
+
+    Args:
+      random_variables: Tuple of `RandomVariable`.
+    """
+    self._random_variables = random_variables
+
+  def __call__(self, event):
+    if isinstance(event, FiniteProductEvent):
+      assert len(event.events) == len(self._random_variables)
+      zipped = list(zip(self._random_variables, event.events))
+      return FiniteProductEvent(
+          [random_variable(sub_event)
+           for random_variable, sub_event in zipped])
+    else:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+  def inverse(self, event):
+    # Specialization for `FiniteProductEvent`; don't need to take all sequences.
+    if isinstance(event, FiniteProductEvent):
+      assert len(event.events) == len(self._random_variables)
+      zipped = list(zip(self._random_variables, event.events))
+      return FiniteProductEvent(tuple(
+          random_variable.inverse(sub_event)
+          for random_variable, sub_event in zipped))
+
+    # Try fallback of mapping each sequence separately.
+    try:
+      all_sequences = event.all_sequences()
+    except AttributeError:
+      raise ValueError('Unhandled event type {}'.format(type(event)))
+
+    mapped = set()
+    for sequence in all_sequences:
+      assert len(sequence) == len(self._random_variables)
+      zipped = list(zip(self._random_variables, sequence))
+      mapped_sequence = FiniteProductEvent(tuple(
+          random_variable.inverse(DiscreteEvent({element}))
+          for random_variable, element in zipped))
+      mapped.update(mapped_sequence.all_sequences())
+    return SequenceEvent(mapped)
diff --git a/mathematics_dataset/mathematics_dataset/util/probability_test.py b/mathematics_dataset/mathematics_dataset/util/probability_test.py
new file mode 100644
index 0000000000000000000000000000000000000000..f474f05ee233631a05b60fbd062b25e33a0b698e
--- /dev/null
+++ b/mathematics_dataset/mathematics_dataset/util/probability_test.py
@@ -0,0 +1,173 @@
+# Copyright 2018 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for mathematics_dataset.util.probability."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+# Dependency imports
+from absl.testing import absltest
+from util import probability
+import sympy
+
+
+class FiniteProductEventTest(absltest.TestCase):
+
+  def testAllSequences(self):
+    event = probability.FiniteProductEvent([probability.DiscreteEvent({1, 2}),
+                                            probability.DiscreteEvent({3})])
+    all_sequences = [i for i in event.all_sequences()]
+    self.assertEqual(all_sequences, [(1, 3), (2, 3)])
+
+
+class CountLevelSetEventTest(absltest.TestCase):
+
+  def testAllSequences(self):
+    event = probability.CountLevelSetEvent({'a': 2, 'b': 4, 'c': 1})
+    all_sequences = event.all_sequences()
+
+    # Number of sequences should be 7! / (4! * 2! * 1!) = 105.
+    self.assertLen(all_sequences, 105)
+    # They should all be unique.
+    self.assertEqual(len(all_sequences), len(set(all_sequences)))
+    # And check contains one correctly generated tuple.
+    self.assertIn(('a', 'b', 'c', 'b', 'b', 'a', 'b'), all_sequences)
+
+
+class DiscreteProbabilitySpaceTest(absltest.TestCase):
+
+  def testBasic(self):
+    space = probability.DiscreteProbabilitySpace({0: 1, 1: 2, 2: 3})
+    p = space.probability(probability.DiscreteEvent([0]))
+    self.assertEqual(p, sympy.Rational(1, 6))
+    p = space.probability(probability.DiscreteEvent([0, 1]))
+    self.assertEqual(p, sympy.Rational(1, 2))
+    p = space.probability(probability.DiscreteEvent([0, 1, 2]))
+    self.assertEqual(p, 1)
+    p = space.probability(probability.DiscreteEvent([0, 1, 2, 3]))
+    self.assertEqual(p, 1)
+    p = space.probability(probability.DiscreteEvent([3]))
+    self.assertEqual(p, 0)
+
+
+class FiniteProductSpaceTest(absltest.TestCase):
+
+  def testProbability_FiniteProductEvent(self):
+    # 5 coin flips of a biased coin with heads prob = 1/3.
+    base_space = probability.DiscreteProbabilitySpace({'h': 1, 't': 2})
+    space = probability.FiniteProductSpace([base_space] * 5)
+
+    heads = probability.DiscreteEvent({'h'})
+    tails = probability.DiscreteEvent({'t'})
+    event = probability.FiniteProductEvent([heads, heads, tails, tails, heads])
+    self.assertEqual(space.probability(event), sympy.Rational(4, 3**5))
+
+  def testProbability_CountLevelSetEvent(self):
+    base_space = probability.DiscreteProbabilitySpace({'a': 2, 'b': 3, 'c': 5})
+    space = probability.FiniteProductSpace([base_space] * 12)
+    event = probability.CountLevelSetEvent({'a': 7, 'b': 2, 'c': 3})
+
+    # Probability should be (12 choose 7 2 3) * p(a)^7 p(b)^2 p(c)^3
+    coeff = 7920
+    p_a = sympy.Rational(1, 5)
+    p_b = sympy.Rational(3, 10)
+    p_c = sympy.Rational(1, 2)
+    self.assertEqual(space.probability(event),
+                     coeff * pow(p_a, 7) * pow(p_b, 2) * pow(p_c, 3))
+
+
+class SampleWithoutReplacementSpaceTest(absltest.TestCase):
+
+  def testBasic(self):
+    space = probability.SampleWithoutReplacementSpace({0: 1, 1: 1}, 2)
+    event_0_0 = probability.FiniteProductEvent(
+        [probability.DiscreteEvent({0}), probability.DiscreteEvent({0})])
+    event_0_1 = probability.FiniteProductEvent(
+        [probability.DiscreteEvent({0}), probability.DiscreteEvent({1})])
+    p_0_0 = space.probability(event_0_0)
+    p_0_1 = space.probability(event_0_1)
+    self.assertEqual(p_0_0, 0)
+    self.assertEqual(p_0_1, sympy.Rational(1, 2))
+
+    space = probability.SampleWithoutReplacementSpace({0: 1, 1: 0}, 1)
+    event_0 = probability.FiniteProductEvent([probability.DiscreteEvent({0})])
+    event_1 = probability.FiniteProductEvent([probability.DiscreteEvent({1})])
+    event_2 = probability.FiniteProductEvent([probability.DiscreteEvent({2})])
+    p_0 = space.probability(event_0)
+    p_1 = space.probability(event_1)
+    p_2 = space.probability(event_2)
+    self.assertEqual(p_0, 1)
+    self.assertEqual(p_1, 0)
+    self.assertEqual(p_2, 0)
+
+
+class DiscreteRandomVariableTest(absltest.TestCase):
+
+  def testCall(self):
+    random_variable = probability.DiscreteRandomVariable({1: 1, 2: 3, 3: 4})
+    forwards = random_variable(probability.DiscreteEvent({1, 3}))
+    self.assertEqual(forwards.values, {1, 4})
+
+  def testInverse(self):
+    random_variable = probability.DiscreteRandomVariable({1: 1, 2: 3, 3: 4})
+    inverse = random_variable.inverse(probability.DiscreteEvent({1, 3}))
+    self.assertEqual(inverse.values, {1, 2})
+
+    random_variable = probability.DiscreteRandomVariable({1: 1, 2: 1})
+    inverse = random_variable.inverse(probability.DiscreteEvent({1, 5}))
+    self.assertEqual(inverse.values, {1, 2})
+
+
+class FiniteProductRandomVariableTest(absltest.TestCase):
+
+  def _random_variable(self):
+    rv1 = probability.DiscreteRandomVariable({1: 'a', 2: 'b', 3: 'c'})
+    rv2 = probability.DiscreteRandomVariable({1: 'x', 2: 'y', 3: 'x'})
+    return probability.FiniteProductRandomVariable((rv1, rv2))
+
+  def testCall_FiniteProductEvent(self):
+    rv = self._random_variable()
+    event1 = probability.DiscreteEvent({1, 2})
+    event2 = probability.DiscreteEvent({1, 3})
+    event = probability.FiniteProductEvent((event1, event2))
+    result = rv(event)
+    self.assertIsInstance(result, probability.FiniteProductEvent)
+    self.assertLen(result.events, 2)
+    self.assertEqual(result.events[0].values, {'a', 'b'})
+    self.assertEqual(result.events[1].values, {'x'})
+
+  def testInverse_FiniteProductEvent(self):
+    rv = self._random_variable()
+    event1 = probability.DiscreteEvent({'a', 'b'})
+    event2 = probability.DiscreteEvent({'x'})
+    event = probability.FiniteProductEvent((event1, event2))
+    result = rv.inverse(event)
+    self.assertIsInstance(result, probability.FiniteProductEvent)
+    self.assertLen(result.events, 2)
+    self.assertEqual(result.events[0].values, {1, 2})
+    self.assertEqual(result.events[1].values, {1, 3})
+
+  def testInverse_CountLevelSetEvent(self):
+    rv = self._random_variable()
+    event = probability.CountLevelSetEvent({'a': 1, 'x': 1})
+    result = rv.inverse(event)
+    sequences = result.all_sequences()
+    self.assertLen(sequences, 2)
+    self.assertEqual(set(sequences), {(1, 1), (1, 3)})
+
+
+if __name__ == '__main__':
+  absltest.main()
diff --git a/mathematics_dataset/setup.py b/mathematics_dataset/setup.py
new file mode 100644
index 0000000000000000000000000000000000000000..150dd3f3599ea68a56767f1c669585a765e5d914
--- /dev/null
+++ b/mathematics_dataset/setup.py
@@ -0,0 +1,66 @@
+# Copyright 2019 DeepMind Technologies Limited.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Module setuptools script."""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from setuptools import find_packages
+from setuptools import setup
+
+description = """A synthetic dataset of school-level mathematics questions.
+
+This dataset code generates mathematical question and answer pairs, from a range
+of question types (such as in arithmetic, algebra, probability, etc), at roughly
+school-level difficulty. This is designed to test the mathematical learning and
+reasoning skills of learning models.
+
+Original paper: Analysing Mathematical Reasoning Abilities of Neural Models
+(Saxton, Grefenstette, Hill, Kohli) (https://openreview.net/pdf?id=H1gR5iR5FX).
+"""
+
+setup(
+    name='mathematics_dataset',
+    version='1.0.1',
+    description='A synthetic dataset of school-level mathematics questions',
+    long_description=description,
+    author='DeepMind',
+    author_email='saxton@google.com',
+    license='Apache License, Version 2.0',
+    keywords='mathematics dataset',
+    url='https://github.com/deepmind/mathematics_dataset',
+    packages=find_packages(),
+    install_requires=[
+        'absl-py>=0.1.0',
+        'numpy>=1.10',
+        'six',
+        'sympy>=1.2',
+    ],
+    classifiers=[
+        'Development Status :: 4 - Beta',
+        'Environment :: Console',
+        'Intended Audience :: Developers',
+        'Intended Audience :: Science/Research',
+        'License :: OSI Approved :: Apache Software License',
+        'Operating System :: POSIX :: Linux',
+        'Programming Language :: Python :: 2.7',
+        'Programming Language :: Python :: 3.4',
+        'Programming Language :: Python :: 3.5',
+        'Programming Language :: Python :: 3.6',
+        'Programming Language :: Python :: 3.7',
+        'Topic :: Scientific/Engineering :: Artificial Intelligence',
+    ],
+)