Jason Alan Fries commited on
Commit
62390ae
1 Parent(s): 9408b89

use sequence tagging formulation of dataset

Browse files
Files changed (1) hide show
  1. blurb.py +12 -19
blurb.py CHANGED
@@ -44,7 +44,6 @@ import re
44
  import pandas
45
  import datasets
46
 
47
- #from .bigbiohub import kb_features
48
  from .bigbiohub import BigBioConfig
49
  from .bigbiohub import Tasks
50
 
@@ -165,7 +164,7 @@ _DESCRIPTION = {
165
 
166
  _HOMEPAGE = "https://microsoft.github.io/BLURB/tasks.html"
167
 
168
- _LICENSE = "MIXED" # Licenses.GPL_3p0
169
 
170
 
171
  _URLs = {
@@ -202,8 +201,7 @@ _BIGBIO_VERSION = "1.0.0"
202
 
203
 
204
  class BlurbDataset(datasets.GeneratorBasedBuilder):
205
- """Source splits for BLURB data (train/val/test) for easy access.
206
- """
207
 
208
  DEFAULT_CONFIG_NAME = "bc5chem"
209
  SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
@@ -253,12 +251,12 @@ class BlurbDataset(datasets.GeneratorBasedBuilder):
253
  {
254
  "id": datasets.Value("string"),
255
  "tokens": datasets.Sequence(datasets.Value("string")),
256
- "word_idx": datasets.Value("int32"),
257
  "type": datasets.Value("string"),
258
- "tags": datasets.Sequence(
259
  datasets.features.ClassLabel(
260
  names=[
261
  "O",
 
262
  "I",
263
  ]
264
  )
@@ -314,7 +312,7 @@ class BlurbDataset(datasets.GeneratorBasedBuilder):
314
  if line.strip() == "":
315
  toks, tags = zip(*tagged)
316
  # transform tags
317
- tags = [t[0] if t[0] in ["I", "O"] else "I" for t in tags]
318
  yield (toks, tags)
319
  tagged = []
320
  continue
@@ -341,16 +339,11 @@ class BlurbDataset(datasets.GeneratorBasedBuilder):
341
 
342
  uid = 0
343
  for item in self._load_iob(filepath):
344
- if len(item) != 2:
345
- print("ERROR", len(item), item, split)
346
- continue
347
  toks, tags = item
348
- for word_idx in range(len(toks)):
349
- yield uid, {
350
- "id": uid,
351
- "tokens": toks,
352
- "type": ner_types[self.config.name],
353
- "tags": tags,
354
- "word_idx": word_idx,
355
- }
356
- uid += 1
 
44
  import pandas
45
  import datasets
46
 
 
47
  from .bigbiohub import BigBioConfig
48
  from .bigbiohub import Tasks
49
 
 
164
 
165
  _HOMEPAGE = "https://microsoft.github.io/BLURB/tasks.html"
166
 
167
+ _LICENSE = "MIXED"
168
 
169
 
170
  _URLs = {
 
201
 
202
 
203
  class BlurbDataset(datasets.GeneratorBasedBuilder):
204
+ """Source splits for BLURB data (train/val/test) for easy access."""
 
205
 
206
  DEFAULT_CONFIG_NAME = "bc5chem"
207
  SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
 
251
  {
252
  "id": datasets.Value("string"),
253
  "tokens": datasets.Sequence(datasets.Value("string")),
 
254
  "type": datasets.Value("string"),
255
+ "ner_tags": datasets.Sequence(
256
  datasets.features.ClassLabel(
257
  names=[
258
  "O",
259
+ "B",
260
  "I",
261
  ]
262
  )
 
312
  if line.strip() == "":
313
  toks, tags = zip(*tagged)
314
  # transform tags
315
+ tags = tags = [t[0] for t in tags]
316
  yield (toks, tags)
317
  tagged = []
318
  continue
 
339
 
340
  uid = 0
341
  for item in self._load_iob(filepath):
 
 
 
342
  toks, tags = item
343
+ yield uid, {
344
+ "id": uid,
345
+ "tokens": toks,
346
+ "type": ner_types[self.config.name],
347
+ "ner_tags": tags,
348
+ }
349
+ uid += 1