ArneBinder
commited on
Commit
•
79740b5
1
Parent(s):
751cc48
from https://github.com/ArneBinder/pie-datasets/pull/138
Browse files- README.md +48 -0
- chemprot.py +273 -0
- requirements.txt +1 -0
README.md
ADDED
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# PIE Dataset Card for "ChemProt"
|
2 |
+
|
3 |
+
This is a [PyTorch-IE](https://github.com/ChristophAlt/pytorch-ie) wrapper for the
|
4 |
+
[ChemProt Huggingface dataset loading script](https://huggingface.co/datasets/bigbio/chemprot).
|
5 |
+
|
6 |
+
## Data Schema
|
7 |
+
|
8 |
+
There are three versions of the dataset supported, `chemprot_full_source`, `chemprot_shared_task_eval_source` and `chemprot_bigbio_kb`.
|
9 |
+
|
10 |
+
#### `ChemprotDocument` for `chemprot_source` and `chemprot_shared_task_eval_source`
|
11 |
+
|
12 |
+
defines following fields:
|
13 |
+
|
14 |
+
- `text` (str)
|
15 |
+
- `id` (str, optional)
|
16 |
+
- `metadata` (dictionary, optional)
|
17 |
+
|
18 |
+
and the following annotation layers:
|
19 |
+
|
20 |
+
- `entities` (annotation type: `LabeledSpan`, target: `text`)
|
21 |
+
- `relations` (annotation type: `BinaryRelation`, target: `entities`)
|
22 |
+
|
23 |
+
#### `ChemprotBigbioDocument` for `chemprot_bigbio_kb`
|
24 |
+
|
25 |
+
defines following fields:
|
26 |
+
|
27 |
+
- `text` (str)
|
28 |
+
- `id` (str, optional)
|
29 |
+
- `metadata` (dictionary, optional)
|
30 |
+
|
31 |
+
and the following annotation layers:
|
32 |
+
|
33 |
+
- `passages` (annotation type: `LabeledSpan`, target: `text`)
|
34 |
+
- `entities` (annotation type: `LabeledSpan`, target: `text`)
|
35 |
+
- `relations` (annotation type: `BinaryRelation`, target: `entities`)
|
36 |
+
|
37 |
+
See [here](https://github.com/ArneBinder/pie-modules/blob/main/src/pie_modules/annotations.py) for the annotation
|
38 |
+
type definitions.
|
39 |
+
|
40 |
+
## Document Converters
|
41 |
+
|
42 |
+
The dataset provides predefined document converters for the following target document types:
|
43 |
+
|
44 |
+
- `pie_modules.documents.TextDocumentWithLabeledSpansAndBinaryRelations` for `ChemprotDocument`
|
45 |
+
- `pie_modules.documents.TextDocumentWithLabeledSpansBinaryRelationsAndLabeledPartitions` for `ChemprotBigbioDocument`
|
46 |
+
|
47 |
+
See [here](https://github.com/ArneBinder/pie-modules/blob/main/src/pie_modules/documents.py) for the document type
|
48 |
+
definitions.
|
chemprot.py
ADDED
@@ -0,0 +1,273 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from dataclasses import dataclass
|
2 |
+
from typing import Any, Dict
|
3 |
+
|
4 |
+
import datasets
|
5 |
+
from pytorch_ie import Document
|
6 |
+
from pytorch_ie.annotations import BinaryRelation, LabeledSpan
|
7 |
+
from pytorch_ie.documents import (
|
8 |
+
AnnotationLayer,
|
9 |
+
TextBasedDocument,
|
10 |
+
TextDocumentWithLabeledSpansAndBinaryRelations,
|
11 |
+
TextDocumentWithLabeledSpansBinaryRelationsAndLabeledPartitions,
|
12 |
+
annotation_field,
|
13 |
+
)
|
14 |
+
|
15 |
+
from pie_datasets import GeneratorBasedBuilder
|
16 |
+
|
17 |
+
|
18 |
+
@dataclass
|
19 |
+
class ChemprotDocument(TextBasedDocument):
|
20 |
+
# used by chemprot_full_source and chemprot_shared_task_eval_source
|
21 |
+
entities: AnnotationLayer[LabeledSpan] = annotation_field(target="text")
|
22 |
+
relations: AnnotationLayer[BinaryRelation] = annotation_field(target="entities")
|
23 |
+
|
24 |
+
|
25 |
+
@dataclass
|
26 |
+
class ChemprotBigbioDocument(TextBasedDocument):
|
27 |
+
passages: AnnotationLayer[LabeledSpan] = annotation_field(target="text")
|
28 |
+
entities: AnnotationLayer[LabeledSpan] = annotation_field(target="text")
|
29 |
+
relations: AnnotationLayer[BinaryRelation] = annotation_field(target="entities")
|
30 |
+
|
31 |
+
|
32 |
+
def example_to_chemprot_doc(example) -> ChemprotDocument:
|
33 |
+
metadata = {"entity_ids": []}
|
34 |
+
id_to_labeled_span: Dict[str, LabeledSpan] = {}
|
35 |
+
|
36 |
+
doc = ChemprotDocument(
|
37 |
+
text=example["text"],
|
38 |
+
id=example["pmid"],
|
39 |
+
metadata=metadata,
|
40 |
+
)
|
41 |
+
|
42 |
+
for idx in range(len(example["entities"]["id"])):
|
43 |
+
labeled_span = LabeledSpan(
|
44 |
+
start=example["entities"]["offsets"][idx][0],
|
45 |
+
end=example["entities"]["offsets"][idx][1],
|
46 |
+
label=example["entities"]["type"][idx],
|
47 |
+
)
|
48 |
+
doc.entities.append(labeled_span)
|
49 |
+
doc.metadata["entity_ids"].append(example["entities"]["id"][idx])
|
50 |
+
id_to_labeled_span[example["entities"]["id"][idx]] = labeled_span
|
51 |
+
|
52 |
+
for idx in range(len(example["relations"]["type"])):
|
53 |
+
doc.relations.append(
|
54 |
+
BinaryRelation(
|
55 |
+
head=id_to_labeled_span[example["relations"]["arg1"][idx]],
|
56 |
+
tail=id_to_labeled_span[example["relations"]["arg2"][idx]],
|
57 |
+
label=example["relations"]["type"][idx],
|
58 |
+
)
|
59 |
+
)
|
60 |
+
|
61 |
+
return doc
|
62 |
+
|
63 |
+
|
64 |
+
def example_to_chemprot_bigbio_doc(example) -> ChemprotBigbioDocument:
|
65 |
+
text = " ".join([" ".join(passage["text"]) for passage in example["passages"]])
|
66 |
+
metadata = {"id": example["id"], "entity_ids": [], "relation_ids": []}
|
67 |
+
id_to_labeled_span: Dict[str, LabeledSpan] = {}
|
68 |
+
|
69 |
+
doc = ChemprotBigbioDocument(
|
70 |
+
text=text,
|
71 |
+
id=example["document_id"],
|
72 |
+
metadata=metadata,
|
73 |
+
)
|
74 |
+
|
75 |
+
for passage in example["passages"]:
|
76 |
+
doc.passages.append(
|
77 |
+
LabeledSpan(
|
78 |
+
start=passage["offsets"][0][0],
|
79 |
+
end=passage["offsets"][0][1],
|
80 |
+
label=passage["type"],
|
81 |
+
)
|
82 |
+
)
|
83 |
+
|
84 |
+
for span in example["entities"]:
|
85 |
+
labeled_span = LabeledSpan(
|
86 |
+
start=span["offsets"][0][0],
|
87 |
+
end=span["offsets"][0][1],
|
88 |
+
label=span["type"],
|
89 |
+
)
|
90 |
+
doc.entities.append(labeled_span)
|
91 |
+
doc.metadata["entity_ids"].append(span["id"])
|
92 |
+
id_to_labeled_span[span["id"]] = labeled_span
|
93 |
+
|
94 |
+
for relation in example["relations"]:
|
95 |
+
doc.relations.append(
|
96 |
+
BinaryRelation(
|
97 |
+
head=id_to_labeled_span[relation["arg1_id"]],
|
98 |
+
tail=id_to_labeled_span[relation["arg2_id"]],
|
99 |
+
label=relation["type"],
|
100 |
+
)
|
101 |
+
)
|
102 |
+
doc.metadata["relation_ids"].append([relation["arg1_id"], relation["arg2_id"]])
|
103 |
+
|
104 |
+
return doc
|
105 |
+
|
106 |
+
|
107 |
+
def chemprot_doc_to_example(doc: ChemprotDocument) -> Dict[str, Any]:
|
108 |
+
entities = {
|
109 |
+
"id": [],
|
110 |
+
"offsets": [],
|
111 |
+
"text": [],
|
112 |
+
"type": [],
|
113 |
+
}
|
114 |
+
relations = {
|
115 |
+
"arg1": [],
|
116 |
+
"arg2": [],
|
117 |
+
"type": [],
|
118 |
+
}
|
119 |
+
|
120 |
+
entity_id2entity = {
|
121 |
+
ent_id: entity for ent_id, entity in zip(doc.metadata["entity_ids"], doc.entities)
|
122 |
+
}
|
123 |
+
|
124 |
+
for entity_id, entity in zip(doc.metadata["entity_ids"], doc.entities):
|
125 |
+
entities["id"].append(entity_id)
|
126 |
+
entities["offsets"].append([entity.start, entity.end])
|
127 |
+
entities["text"].append(doc.text[entity.start : entity.end])
|
128 |
+
entities["type"].append(entity.label)
|
129 |
+
|
130 |
+
if entity in entity_id2entity:
|
131 |
+
raise ValueError("Entity already exists in entity_id2entity")
|
132 |
+
|
133 |
+
entity_id2entity[entity] = entity_id
|
134 |
+
|
135 |
+
for relation in doc.relations:
|
136 |
+
relations["arg1"].append(entity_id2entity[relation.head])
|
137 |
+
relations["arg2"].append(entity_id2entity[relation.tail])
|
138 |
+
relations["type"].append(relation.label)
|
139 |
+
|
140 |
+
return {
|
141 |
+
"text": doc.text,
|
142 |
+
"pmid": doc.id,
|
143 |
+
"entities": entities,
|
144 |
+
"relations": relations,
|
145 |
+
}
|
146 |
+
|
147 |
+
|
148 |
+
def chemprot_bigbio_doc_to_example(doc: ChemprotBigbioDocument) -> Dict[str, Any]:
|
149 |
+
id = int(doc.metadata["id"])
|
150 |
+
passages = []
|
151 |
+
entities = []
|
152 |
+
relations = []
|
153 |
+
|
154 |
+
entity_id2entity = {
|
155 |
+
ent_id: entity for ent_id, entity in zip(doc.metadata["entity_ids"], doc.entities)
|
156 |
+
}
|
157 |
+
|
158 |
+
for passage in doc.passages:
|
159 |
+
id += 1
|
160 |
+
passages.append(
|
161 |
+
{
|
162 |
+
"id": str(id),
|
163 |
+
"offsets": [[passage.start, passage.end]],
|
164 |
+
"text": [doc.text[passage.start : passage.end]],
|
165 |
+
"type": passage.label,
|
166 |
+
}
|
167 |
+
)
|
168 |
+
|
169 |
+
entity2entity_id = dict()
|
170 |
+
|
171 |
+
for entity_id, entity in zip(doc.metadata["entity_ids"], doc.entities):
|
172 |
+
id += 1
|
173 |
+
entities.append(
|
174 |
+
{
|
175 |
+
"id": entity_id, # entity_id = str(id)
|
176 |
+
"normalized": [],
|
177 |
+
"offsets": [[entity.start, entity.end]],
|
178 |
+
"text": [doc.text[entity.start : entity.end]],
|
179 |
+
"type": entity.label,
|
180 |
+
}
|
181 |
+
)
|
182 |
+
if entity in entity_id2entity:
|
183 |
+
raise ValueError("Entity already exists in entity_id2entity")
|
184 |
+
|
185 |
+
entity2entity_id[entity] = entity_id
|
186 |
+
|
187 |
+
for relation in doc.relations:
|
188 |
+
id += 1
|
189 |
+
relations.append(
|
190 |
+
{
|
191 |
+
"id": str(id), # save in metadata?
|
192 |
+
"arg1_id": entity2entity_id[relation.head],
|
193 |
+
"arg2_id": entity2entity_id[relation.tail],
|
194 |
+
"type": relation.label,
|
195 |
+
"normalized": [],
|
196 |
+
}
|
197 |
+
)
|
198 |
+
|
199 |
+
return {
|
200 |
+
"id": doc.metadata["id"],
|
201 |
+
"document_id": doc.id,
|
202 |
+
"passages": passages,
|
203 |
+
"entities": entities,
|
204 |
+
"events": [],
|
205 |
+
"coreferences": [],
|
206 |
+
"relations": relations,
|
207 |
+
}
|
208 |
+
|
209 |
+
|
210 |
+
class Chemprot(GeneratorBasedBuilder):
|
211 |
+
DOCUMENT_TYPES = { # Note ChemprotDocument is used twice
|
212 |
+
"chemprot_full_source": ChemprotDocument,
|
213 |
+
"chemprot_bigbio_kb": ChemprotBigbioDocument,
|
214 |
+
"chemprot_shared_task_eval_source": ChemprotDocument,
|
215 |
+
}
|
216 |
+
|
217 |
+
BASE_DATASET_PATH = "bigbio/chemprot"
|
218 |
+
BASE_DATASET_REVISION = "86afccf3ccc614f817a7fad0692bf62fbc5ce469"
|
219 |
+
|
220 |
+
BUILDER_CONFIGS = [
|
221 |
+
datasets.BuilderConfig(
|
222 |
+
name="chemprot_full_source",
|
223 |
+
version=datasets.Version("1.0.0"),
|
224 |
+
description="ChemProt full source version",
|
225 |
+
),
|
226 |
+
datasets.BuilderConfig(
|
227 |
+
name="chemprot_bigbio_kb",
|
228 |
+
version=datasets.Version("1.0.0"),
|
229 |
+
description="ChemProt BigBio kb version",
|
230 |
+
),
|
231 |
+
datasets.BuilderConfig(
|
232 |
+
name="chemprot_shared_task_eval_source",
|
233 |
+
version=datasets.Version("1.0.0"),
|
234 |
+
description="ChemProt shared task eval source version",
|
235 |
+
),
|
236 |
+
]
|
237 |
+
|
238 |
+
@property
|
239 |
+
def document_converters(self):
|
240 |
+
if (
|
241 |
+
self.config.name == "chemprot_full_source"
|
242 |
+
or self.config.name == "chemprot_shared_task_eval_source"
|
243 |
+
):
|
244 |
+
return {
|
245 |
+
TextDocumentWithLabeledSpansAndBinaryRelations: {
|
246 |
+
"entities": "labeled_spans",
|
247 |
+
"relations": "binary_relations",
|
248 |
+
}
|
249 |
+
}
|
250 |
+
elif self.config.name == "chemprot_bigbio_kb":
|
251 |
+
return {
|
252 |
+
TextDocumentWithLabeledSpansBinaryRelationsAndLabeledPartitions: {
|
253 |
+
"passages": "labeled_partitions",
|
254 |
+
"entities": "labeled_spans",
|
255 |
+
"relations": "binary_relations",
|
256 |
+
}
|
257 |
+
}
|
258 |
+
else:
|
259 |
+
raise ValueError(f"Unknown dataset name: {self.config.name}")
|
260 |
+
|
261 |
+
def _generate_document(self, example, **kwargs):
|
262 |
+
if self.config.name == "chemprot_bigbio_kb":
|
263 |
+
return example_to_chemprot_bigbio_doc(example)
|
264 |
+
else:
|
265 |
+
return example_to_chemprot_doc(example)
|
266 |
+
|
267 |
+
def _generate_example(self, document: Document, **kwargs) -> Dict[str, Any]:
|
268 |
+
if isinstance(document, ChemprotBigbioDocument):
|
269 |
+
return chemprot_bigbio_doc_to_example(document)
|
270 |
+
elif isinstance(document, ChemprotDocument):
|
271 |
+
return chemprot_doc_to_example(document)
|
272 |
+
else:
|
273 |
+
raise ValueError(f"Unknown document type: {type(document)}")
|
requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
pie-datasets>=0.6.0,<0.11.0
|