Convert dataset to Parquet (#2)

- Convert dataset to Parquet (ffd96bdfa1890b305a8f6d10b8bad0aadac5fc1c)
- Delete loading script (1251ea30cab7d811a89a9810b280dd0ab2c9f9fd)
- Delete legacy dataset_infos.json (c9e8b5f6725ede38f0fc0c9a36cb85d46a9fe168)

README.md

@@ -1,5 +1,4 @@
 ---
-pretty_name: DROP
 annotations_creators:
 - crowdsourced
 language_creators:
@@ -21,6 +20,7 @@ task_ids:
 - extractive-qa
 - abstractive-qa
 paperswithcode_id: drop
+pretty_name: DROP
 dataset_info:
   features:
   - name: section_id
@@ -39,13 +39,20 @@ dataset_info:
       dtype: string
   splits:
   - name: train
-    num_bytes: 105572762
+    num_bytes: 105572506
     num_examples: 77400
   - name: validation
-    num_bytes: 11737787
+    num_bytes: 11737755
     num_examples: 9535
-  download_size: 8308692
-  dataset_size: 117310549
+  download_size: 11538387
+  dataset_size: 117310261
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train-*
+  - split: validation
+    path: data/validation-*
 ---
 
 # Dataset Card for "drop"

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:abb5e578c2156a61f83b56c066a922ec1c7c5140638a3f0f2a7c348fafe1cb35
+size 10333127

data/validation-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f9a3bdbb1b5909abfa25cbab693f89f47568c98e6e03473500d604f044c8f68
+size 1205260

dataset_infos.json

@@ -1 +0,0 @@
-{"default": {"description": "DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs.\n. DROP is a crowdsourced, adversarially-created, 96k-question benchmark, in which a system must resolve references in a\nquestion, perhaps to multiple input positions, and perform discrete operations over them (such as addition, counting, or\n sorting). These operations require a much more comprehensive understanding of the content of paragraphs than what was\n necessary for prior datasets.\n", "citation": "@inproceedings{Dua2019DROP,\n  author={Dheeru Dua and Yizhong Wang and Pradeep Dasigi and Gabriel Stanovsky and Sameer Singh and Matt Gardner},\n  title={DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs},\n  booktitle={Proc. of NAACL},\n  year={2019}\n}\n", "homepage": "https://allennlp.org/drop", "license": "", "features": {"section_id": {"dtype": "string", "id": null, "_type": "Value"}, "query_id": {"dtype": "string", "id": null, "_type": "Value"}, "passage": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "answers_spans": {"feature": {"spans": {"dtype": "string", "id": null, "_type": "Value"}, "types": {"dtype": "string", "id": null, "_type": "Value"}}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "drop", "config_name": "default", "version": {"version_str": "0.1.0", "description": null, "major": 0, "minor": 1, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 105572762, "num_examples": 77400, "dataset_name": "drop"}, "validation": {"name": "validation", "num_bytes": 11737787, "num_examples": 9535, "dataset_name": "drop"}}, "download_checksums": {"https://s3-us-west-2.amazonaws.com/allennlp/datasets/drop/drop_dataset.zip": {"num_bytes": 8308692, "checksum": "39d2278a29fd729de301b111a45f434c24834f40df8f4ff116d864589e3249d6"}}, "download_size": 8308692, "post_processing_size": null, "dataset_size": 117310549, "size_in_bytes": 125619241}}

drop.py

@@ -1,202 +0,0 @@
-"""TODO(drop): Add a description here."""
-
-
-import json
-import os
-
-import datasets
-
-
-_CITATION = """\
-@inproceedings{Dua2019DROP,
-  author={Dheeru Dua and Yizhong Wang and Pradeep Dasigi and Gabriel Stanovsky and Sameer Singh and Matt Gardner},
-  title={DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs},
-  booktitle={Proc. of NAACL},
-  year={2019}
-}
-"""
-
-_DESCRIPTION = """\
-DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs.
-. DROP is a crowdsourced, adversarially-created, 96k-question benchmark, in which a system must resolve references in a
-question, perhaps to multiple input positions, and perform discrete operations over them (such as addition, counting, or
- sorting). These operations require a much more comprehensive understanding of the content of paragraphs than what was
- necessary for prior datasets.
-"""
-_URL = "https://s3-us-west-2.amazonaws.com/allennlp/datasets/drop/drop_dataset.zip"
-
-
-class AnswerParsingError(Exception):
-    pass
-
-
-class DropDateObject:
-    """
-    Custom parser for date answers in DROP.
-    A date answer is a dict <date> with at least one of day|month|year.
-
-    Example: date == {
-        'day': '9',
-        'month': 'March',
-        'year': '2021'
-    }
-
-    This dict is parsed and flattend to '{day} {month} {year}', not including
-    blank values.
-
-    Example: str(DropDateObject(date)) == '9 March 2021'
-    """
-
-    def __init__(self, dict_date):
-        self.year = dict_date.get("year", "")
-        self.month = dict_date.get("month", "")
-        self.day = dict_date.get("day", "")
-
-    def __iter__(self):
-        yield from [self.day, self.month, self.year]
-
-    def __bool__(self):
-        return any(self)
-
-    def __repr__(self):
-        return " ".join(self).strip()
-
-
-class Drop(datasets.GeneratorBasedBuilder):
-    """TODO(drop): Short description of my dataset."""
-
-    # TODO(drop): Set up version.
-    VERSION = datasets.Version("0.1.0")
-
-    def _info(self):
-        # TODO(drop): Specifies the datasets.DatasetInfo object
-        return datasets.DatasetInfo(
-            # This is the description that will appear on the datasets page.
-            description=_DESCRIPTION,
-            # datasets.features.FeatureConnectors
-            features=datasets.Features(
-                {
-                    "section_id": datasets.Value("string"),
-                    "query_id": datasets.Value("string"),
-                    "passage": datasets.Value("string"),
-                    "question": datasets.Value("string"),
-                    "answers_spans": datasets.features.Sequence(
-                        {"spans": datasets.Value("string"), "types": datasets.Value("string")}
-                    )
-                    # These are the features of your dataset like images, labels ...
-                }
-            ),
-            # If there's a common (input, target) tuple from the features,
-            # specify them here. They'll be used if as_supervised=True in
-            # builder.as_dataset.
-            supervised_keys=None,
-            # Homepage of the dataset for documentation
-            homepage="https://allennlp.org/drop",
-            citation=_CITATION,
-        )
-
-    def _split_generators(self, dl_manager):
-        """Returns SplitGenerators."""
-        # TODO(drop): Downloads the data and defines the splits
-        # dl_manager is a datasets.download.DownloadManager that can be used to
-        # download and extract URLs
-        dl_dir = dl_manager.download_and_extract(_URL)
-        data_dir = os.path.join(dl_dir, "drop_dataset")
-        return [
-            datasets.SplitGenerator(
-                name=datasets.Split.TRAIN,
-                # These kwargs will be passed to _generate_examples
-                gen_kwargs={"filepath": os.path.join(data_dir, "drop_dataset_train.json"), "split": "train"},
-            ),
-            datasets.SplitGenerator(
-                name=datasets.Split.VALIDATION,
-                # These kwargs will be passed to _generate_examples
-                gen_kwargs={"filepath": os.path.join(data_dir, "drop_dataset_dev.json"), "split": "validation"},
-            ),
-        ]
-
-    def _generate_examples(self, filepath, split):
-        """Yields examples."""
-        # TODO(drop): Yields (key, example) tuples from the dataset
-        with open(filepath, mode="r", encoding="utf-8") as f:
-            data = json.load(f)
-            id_ = 0
-            for i, (section_id, section) in enumerate(data.items()):
-                for j, qa in enumerate(section["qa_pairs"]):
-
-                    example = {
-                        "section_id": section_id,
-                        "query_id": qa["query_id"],
-                        "passage": section["passage"],
-                        "question": qa["question"],
-                    }
-
-                    if split == "train":
-                        answers = [qa["answer"]]
-                    else:
-                        answers = qa["validated_answers"]
-
-                    try:
-                        example["answers_spans"] = self.build_answers(answers)
-                        yield id_, example
-                        id_ += 1
-                    except AnswerParsingError:
-                        # This is expected for 9 examples of train
-                        # and 1 of validation.
-                        continue
-
-    @staticmethod
-    def _raise(message):
-        """
-        Raise a custom AnswerParsingError, to be sure to only catch our own
-        errors. Messages are irrelavant for this script, but are written to
-        ease understanding the code.
-        """
-        raise AnswerParsingError(message)
-
-    def build_answers(self, answers):
-
-        returned_answers = {
-            "spans": list(),
-            "types": list(),
-        }
-        for answer in answers:
-            date = DropDateObject(answer["date"])
-
-            if answer["number"] != "":
-                # sanity checks
-                if date:
-                    self._raise("This answer is both number and date!")
-                if len(answer["spans"]):
-                    self._raise("This answer is both number and text!")
-
-                returned_answers["spans"].append(answer["number"])
-                returned_answers["types"].append("number")
-
-            elif date:
-                # sanity check
-                if len(answer["spans"]):
-                    self._raise("This answer is both date and text!")
-
-                returned_answers["spans"].append(str(date))
-                returned_answers["types"].append("date")
-
-            # won't triger if len(answer['spans']) == 0
-            for span in answer["spans"]:
-                # sanity checks
-                if answer["number"] != "":
-                    self._raise("This answer is both text and number!")
-                if date:
-                    self._raise("This answer is both text and date!")
-
-                returned_answers["spans"].append(span)
-                returned_answers["types"].append("span")
-
-        # sanity check
-        _len = len(returned_answers["spans"])
-        if not _len:
-            self._raise("Empty answer.")
-        if any(len(l) != _len for _, l in returned_answers.items()):
-            self._raise("Something went wrong while parsing answer values/types")
-
-        return returned_answers