Datasets:

Modalities:
Text
Languages:
English
Size:
< 1K
Libraries:
Datasets
License:
gabrielaltay commited on
Commit
ca22192
·
1 Parent(s): a91125c

upload hubscripts/pdr_hub.py to hub from bigbio repo

Browse files
Files changed (1) hide show
  1. pdr.py +462 -0
pdr.py ADDED
@@ -0,0 +1,462 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+ """
16
+ The corpus of plant-disease relation consists of plants and diseases and their relation to PubMed abstract.
17
+ The corpus consists of about 2400 plant and disease entities and 300 annotated relations from 179 abstracts.
18
+
19
+ The big-bio and source version of this script are made by merging the 2 provided annotations on locations they intersected.
20
+ Both annotations (1, 2) are provided as separate source schemas.
21
+ """
22
+ from collections import defaultdict
23
+ from pathlib import Path
24
+ from typing import Dict, Iterator, Optional, Tuple
25
+
26
+ import datasets
27
+
28
+ from .bigbiohub import
29
+ from .bigbiohub import BigBioConfig
30
+ from .bigbiohub import Tasks
31
+
32
+ _LANGUAGES = ['English']
33
+ _PUBMED = True
34
+ _LOCAL = False
35
+ _CITATION = """\
36
+ @article{kim2019corpus,
37
+ title={A corpus of plant--disease relations in the biomedical domain},
38
+ author={Kim, Baeksoo and Choi, Wonjun and Lee, Hyunju},
39
+ journal={PLoS One},
40
+ volume={14},
41
+ number={8},
42
+ pages={e0221582},
43
+ year={2019},
44
+ publisher={Public Library of Science San Francisco, CA USA}
45
+ }
46
+ """
47
+
48
+ _DATASETNAME = "pdr"
49
+ _DISPLAYNAME = "PDR"
50
+
51
+ _DESCRIPTION = """
52
+ The corpus of plant-disease relation consists of plants and diseases and their relation to PubMed abstract.
53
+ The corpus consists of about 2400 plant and disease entities and 300 annotated relations from 179 abstracts.
54
+ """
55
+
56
+ _HOMEPAGE = "http://gcancer.org/pdr/"
57
+ _LICENSE = 'License information unavailable'
58
+ _URLS = {_DATASETNAME: "http://gcancer.org/pdr/Plant-Disease_Corpus.tar.gz"}
59
+
60
+ _SUPPORTED_TASKS = [
61
+ Tasks.NAMED_ENTITY_RECOGNITION,
62
+ # Tasks.RELATION_EXTRACTION,
63
+ Tasks.EVENT_EXTRACTION,
64
+ Tasks.COREFERENCE_RESOLUTION,
65
+ ]
66
+
67
+ _SOURCE_VERSION = "1.0.0"
68
+ _BIGBIO_VERSION = "1.0.0"
69
+
70
+
71
+ class PDRDataset(datasets.GeneratorBasedBuilder):
72
+ """The corpus of plant-disease relation consists of plants and diseases and their relation to PubMed abstract"""
73
+
74
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
75
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
76
+
77
+ BUILDER_CONFIGS = [
78
+ BigBioConfig(
79
+ name="pdr_annotator1_source",
80
+ version=SOURCE_VERSION,
81
+ description="PDR annotator 1 source schema",
82
+ schema="source",
83
+ subset_id="pdr_annotator1",
84
+ ),
85
+ BigBioConfig(
86
+ name="pdr_annotator2_source",
87
+ version=SOURCE_VERSION,
88
+ description="PDR annotator 2 source schema",
89
+ schema="source",
90
+ subset_id="pdr_annotator2",
91
+ ),
92
+ BigBioConfig(
93
+ name="pdr_source",
94
+ version=SOURCE_VERSION,
95
+ description="PDR source schema",
96
+ schema="source",
97
+ subset_id="pdr",
98
+ ),
99
+ BigBioConfig(
100
+ name="pdr_bigbio_kb",
101
+ version=BIGBIO_VERSION,
102
+ description="PDR BigBio schema",
103
+ schema="bigbio_kb",
104
+ subset_id="pdr",
105
+ ),
106
+ ]
107
+
108
+ DEFAULT_CONFIG_NAME = "pdr_source"
109
+
110
+ def _info(self):
111
+ if self.config.schema == "source":
112
+ features = datasets.Features(
113
+ {
114
+ "document_id": datasets.Value("string"),
115
+ "text": datasets.Value("string"),
116
+ "entities": [
117
+ {
118
+ "id": datasets.Value("string"),
119
+ "type": datasets.Value("string"),
120
+ "offsets": datasets.Sequence([datasets.Value("int32")]),
121
+ "text": datasets.Sequence(datasets.Value("string")),
122
+ "normalized": [
123
+ {
124
+ "db_name": datasets.Value("string"),
125
+ "db_id": datasets.Value("string"),
126
+ }
127
+ ],
128
+ }
129
+ ],
130
+ "relations": [
131
+ {
132
+ "id": datasets.Value("string"),
133
+ "type": datasets.Value("string"),
134
+ "arg1_id": datasets.Value("string"),
135
+ "arg2_id": datasets.Value("string"),
136
+ "normalized": [
137
+ {
138
+ "db_name": datasets.Value("string"),
139
+ "db_id": datasets.Value("string"),
140
+ }
141
+ ],
142
+ }
143
+ ],
144
+ "events": [
145
+ {
146
+ "id": datasets.Value("string"),
147
+ "type": datasets.Value("string"),
148
+ # refers to the text_bound_annotation of the trigger
149
+ "trigger": {
150
+ "text": datasets.Sequence(datasets.Value("string")),
151
+ "offsets": datasets.Sequence([datasets.Value("int32")]),
152
+ },
153
+ "arguments": [
154
+ {
155
+ "role": datasets.Value("string"),
156
+ "ref_id": datasets.Value("string"),
157
+ }
158
+ ],
159
+ }
160
+ ],
161
+ "coreferences": [
162
+ {
163
+ "id": datasets.Value("string"),
164
+ "entity_ids": datasets.Sequence(datasets.Value("string")),
165
+ }
166
+ ],
167
+ },
168
+ )
169
+
170
+ elif self.config.schema == "bigbio_kb":
171
+ features = kb_features
172
+
173
+ return datasets.DatasetInfo(
174
+ description=_DESCRIPTION,
175
+ features=features,
176
+ homepage=_HOMEPAGE,
177
+ license=str(_LICENSE),
178
+ citation=_CITATION,
179
+ )
180
+
181
+ def _split_generators(self, dl_manager):
182
+ urls = _URLS[_DATASETNAME]
183
+ data_dir = Path(dl_manager.download_and_extract(urls))
184
+ data_dir = data_dir / "Plant-Disease_Corpus"
185
+
186
+ return [
187
+ datasets.SplitGenerator(
188
+ name=datasets.Split.TRAIN,
189
+ gen_kwargs={"data_dir": data_dir},
190
+ )
191
+ ]
192
+
193
+ def _generate_examples(self, data_dir: Path) -> Iterator[Tuple[str, Dict]]:
194
+ if self.config.schema == "source":
195
+ for file in data_dir.iterdir():
196
+ if not str(file).endswith(".txt"):
197
+ continue
198
+
199
+ if self.config.subset_id == "pdr_annotator1":
200
+ # Provide annotations of annotator 1
201
+ example = parsing.parse_brat_file(file, [".ann"])
202
+ example = parsing.brat_parse_to_bigbio_kb(example)
203
+
204
+ elif self.config.subset_id == "pdr_annotator2":
205
+ # Provide annotations of annotator 2
206
+ example = parsing.parse_brat_file(file, [".ann2"])
207
+ example = parsing.brat_parse_to_bigbio_kb(example)
208
+
209
+ elif self.config.subset_id == "pdr":
210
+ # Provide merged version of annotator 1 and 2
211
+ annotator1 = parsing.parse_brat_file(file, [".ann"])
212
+ annotator1 = parsing.brat_parse_to_bigbio_kb(annotator1)
213
+
214
+ annotator2 = parsing.parse_brat_file(file, [".ann2"])
215
+ annotator2 = parsing.brat_parse_to_bigbio_kb(annotator2)
216
+
217
+ example = self._merge_annotations_by_intersection(
218
+ file, annotator1, annotator2
219
+ )
220
+
221
+ example["text"] = example["passages"][0]["text"][0]
222
+ example.pop("id", None)
223
+ example.pop("passages", None)
224
+
225
+ yield example["document_id"], example
226
+
227
+ elif self.config.schema == "bigbio_kb":
228
+ for file in data_dir.iterdir():
229
+ if not str(file).endswith(".txt"):
230
+ continue
231
+
232
+ annotator1 = parsing.parse_brat_file(file, [".ann"])
233
+ annotator1 = parsing.brat_parse_to_bigbio_kb(annotator1)
234
+
235
+ annotator2 = parsing.parse_brat_file(file, [".ann2"])
236
+ annotator2 = parsing.brat_parse_to_bigbio_kb(annotator2)
237
+
238
+ merged_annotation = self._merge_annotations_by_intersection(
239
+ file, annotator1, annotator2
240
+ )
241
+ merged_annotation["id"] = merged_annotation["document_id"]
242
+
243
+ yield merged_annotation["id"], merged_annotation
244
+
245
+ def _merge_annotations_by_intersection(
246
+ self, file: Path, example_ann1: Dict, example_ann2: Dict
247
+ ) -> Dict:
248
+ """
249
+ Merges the two given examples by only keeping annotations on which both annotators agree.
250
+ """
251
+ id_prefix = str(file.stem) + "_"
252
+
253
+ # Mapping entity identifiers from annotator 1 / 2 to merged entity ids
254
+ a1_entity_to_merged_entity = {}
255
+ a2_entity_to_merged_entity = {}
256
+ merged_entities = []
257
+
258
+ # 1. Find all common entities, i.e. both annotators agree on same type and their offsets overlap
259
+ entity_id = 1
260
+ for entity1 in example_ann1["entities"]:
261
+ for entity2 in example_ann2["entities"]:
262
+ if (
263
+ self._overlaps(entity1, entity2)
264
+ and entity1["type"] == entity2["type"]
265
+ ):
266
+ text_entity1 = "".join(entity1["text"])
267
+ text_entity2 = "".join(entity2["text"])
268
+
269
+ longer_entity = (
270
+ entity1 if len(text_entity1) > len(text_entity2) else entity2
271
+ )
272
+ merged_entity_id = id_prefix + f"E{entity_id}"
273
+ entity_id += 1
274
+
275
+ merged_entity = longer_entity.copy()
276
+ merged_entity["id"] = merged_entity_id
277
+ merged_entity["normalized"] = []
278
+ merged_entities.append(merged_entity)
279
+
280
+ a1_entity_to_merged_entity[entity1["id"]] = merged_entity_id
281
+ a2_entity_to_merged_entity[entity2["id"]] = merged_entity_id
282
+ break
283
+
284
+ # Find all relations the two annotators agree on
285
+ relations_ann1 = self._map_relations(example_ann1, a1_entity_to_merged_entity)
286
+ relations_ann2 = self._map_relations(example_ann2, a2_entity_to_merged_entity)
287
+ relations = []
288
+ relation_id = 1
289
+
290
+ for rel_type, relations_1 in relations_ann1.items():
291
+ relations_2 = relations_ann2[rel_type]
292
+
293
+ for relation_pair_1 in relations_1:
294
+ for relation_pair_2 in relations_2:
295
+ if relation_pair_1 == relation_pair_2:
296
+ relations.append(
297
+ {
298
+ "id": id_prefix + f"R{relation_id}",
299
+ "type": rel_type,
300
+ "arg1_id": relation_pair_1[0],
301
+ "arg2_id": relation_pair_1[1],
302
+ "normalized": [],
303
+ }
304
+ )
305
+ relation_id += 1
306
+ break
307
+
308
+ # Find all events the two annotators agree on
309
+ events_ann1 = self._map_events(example_ann1, a1_entity_to_merged_entity)
310
+ events_ann2 = self._map_events(example_ann2, a2_entity_to_merged_entity)
311
+ events = []
312
+ event_id = 1
313
+
314
+ for event_type, events_1 in events_ann1.items():
315
+ events_2 = events_ann2[event_type]
316
+
317
+ for (trigger1, theme1, cause1) in events_1:
318
+ for (trigger2, theme2, cause2) in events_2:
319
+ if (
320
+ theme1 == theme2
321
+ and cause1 == cause2
322
+ and self._overlaps(trigger1, trigger2)
323
+ ):
324
+ trigger1_text = "".join(trigger1["text"])
325
+ trigger2_text = "".join(trigger2["text"])
326
+
327
+ longer_trigger = (
328
+ trigger1
329
+ if len(trigger1_text) >= len(trigger2_text)
330
+ else trigger2
331
+ )
332
+ events.append(
333
+ {
334
+ "id": id_prefix + f"T{event_id}",
335
+ "type": event_type,
336
+ "trigger": longer_trigger,
337
+ "arguments": [
338
+ {"role": "Theme", "ref_id": theme1},
339
+ {"role": "Cause", "ref_id": cause1},
340
+ ],
341
+ }
342
+ )
343
+ event_id += 1
344
+ break
345
+
346
+ # Find all coreferences the annotators agree on
347
+ coferences_ann1 = self._map_coreferences(
348
+ example_ann1, a1_entity_to_merged_entity
349
+ )
350
+ coferences_ann2 = self._map_coreferences(
351
+ example_ann2, a2_entity_to_merged_entity
352
+ )
353
+ coreferences = []
354
+ coreference_id = 1
355
+
356
+ for _, entity_ids1 in coferences_ann1.items():
357
+ for _, entity_ids2 in coferences_ann2.items():
358
+ if entity_ids1.intersection(entity_ids2) == entity_ids1.union(
359
+ entity_ids2
360
+ ):
361
+ coreferences.append(
362
+ {
363
+ "id": id_prefix + f"CO{coreference_id}",
364
+ "entity_ids": list(entity_ids1),
365
+ }
366
+ )
367
+ coreference_id += 1
368
+
369
+ merged_example = example_ann1.copy()
370
+ merged_example["entities"] = merged_entities
371
+ merged_example["relations"] = relations
372
+ merged_example["events"] = events
373
+ merged_example["coreferences"] = coreferences
374
+
375
+ return merged_example
376
+
377
+ def _map_relations(self, example: Dict, entity_id_mapping: Dict) -> Dict:
378
+ """
379
+ Maps the all relations of the given example to their merged entity identifiers
380
+ (if existent)
381
+ """
382
+ relation_map = defaultdict(list)
383
+
384
+ for relation in example["relations"]:
385
+ arg1_id = relation["arg1_id"]
386
+ arg2_id = relation["arg2_id"]
387
+
388
+ # Are both entities also in the merged version?
389
+ if arg1_id not in entity_id_mapping or arg2_id not in entity_id_mapping:
390
+ continue
391
+
392
+ com_arg1_id = entity_id_mapping[arg1_id]
393
+ com_arg2_id = entity_id_mapping[arg2_id]
394
+
395
+ relation_map[relation["type"]].append((com_arg1_id, com_arg2_id))
396
+
397
+ return relation_map
398
+
399
+ def _map_events(self, example: Dict, entity_id_mapping: Dict) -> Dict:
400
+ """
401
+ Maps the all events of the given example to their merged entity identifiers
402
+ (if existent)
403
+ """
404
+ event_map = defaultdict(list)
405
+
406
+ for event in example["events"]:
407
+ theme_id = self._get_event_argument(event, "Theme")
408
+ cause_id = self._get_event_argument(event, "Cause")
409
+
410
+ if theme_id not in entity_id_mapping or cause_id not in entity_id_mapping:
411
+ continue
412
+
413
+ common_theme_id = entity_id_mapping[theme_id]
414
+ common_cause_id = entity_id_mapping[cause_id]
415
+
416
+ event_map[event["type"]].append(
417
+ (event["trigger"], common_theme_id, common_cause_id)
418
+ )
419
+
420
+ return event_map
421
+
422
+ def _map_coreferences(self, annotation: Dict, entity_mapping: Dict) -> Dict:
423
+ """
424
+ Maps the all coreferences of the given example to their merged entity identifiers
425
+ (if existent)
426
+ """
427
+ id_to_corefs = defaultdict(set)
428
+ for coreference in annotation["coreferences"]:
429
+ entity_ids = set(
430
+ [
431
+ entity_mapping[id]
432
+ for id in coreference["entity_ids"]
433
+ if id in entity_mapping
434
+ ]
435
+ )
436
+
437
+ # Are both id's also in the merged version?
438
+ if len(entity_ids) > 1:
439
+ id_to_corefs[coreference["id"]] = entity_ids
440
+
441
+ return id_to_corefs
442
+
443
+ def _overlaps(self, annotation1: Dict, annotation2: Dict) -> bool:
444
+ """
445
+ Checks whether the offsets of the two given annotations overlap.
446
+ """
447
+ for (start1, end1) in annotation1["offsets"]:
448
+ for (start2, end2) in annotation2["offsets"]:
449
+ if (start2 <= start1 <= end2) or (start2 <= end1 <= end2):
450
+ return True
451
+
452
+ return False
453
+
454
+ def _get_event_argument(self, event: Dict, role: str) -> Optional[str]:
455
+ """
456
+ Returns the argument with the given role from the given event annotation.
457
+ """
458
+ for argument in event["arguments"]:
459
+ if argument["role"] == role:
460
+ return argument["ref_id"]
461
+
462
+ return None