Datasets:

Tasks:
Other
Modalities:
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:
VictorSanh commited on
Commit
b6b3cf8
1 Parent(s): 1688659
Files changed (1) hide show
  1. P3.py +4 -10
P3.py CHANGED
@@ -78,7 +78,7 @@ _DATA_PATH = "data"
78
  # )
79
  # return ds
80
 
81
- def load_cached_task(features_file, tfrecord, split):
82
  # # TODO(Victor): this info.*.json is actually done twice... -> factorize
83
  # with tf.io.gfile.GFile(os.path.join(cache_dir, f"info.{split}.json")) as f:
84
  with tf.io.gfile.GFile(features_file) as f:
@@ -100,9 +100,6 @@ def load_cached_task(features_file, tfrecord, split):
100
  feat: _feature_config(**desc) for feat, desc in features.items()
101
  }
102
 
103
- # tfrecords = os.path.join(
104
- # cache_dir, f"{split}.tfrecord-*-of-*{split_info['num_shards']}"
105
- # )
106
  ds = tf.data.TFRecordDataset(tf.io.gfile.glob([tfrecord]))
107
  ds = ds.map(
108
  lambda pb: tf.io.parse_single_example(pb, feature_description),
@@ -116,6 +113,7 @@ def load_cached_task(features_file, tfrecord, split):
116
  )
117
  return ds
118
 
 
119
  def find_task_splits_and_features():
120
  """Find the available tasks under ./data and their available splits and features."""
121
  task_and_their_splits = defaultdict(dict)
@@ -239,7 +237,6 @@ class P3(datasets.GeneratorBasedBuilder):
239
  gen_kwargs={
240
  "features_file": data_dir[task_name][split_name]["features_file"],
241
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
242
- "split": split_name,
243
  }
244
  )
245
  )
@@ -251,7 +248,6 @@ class P3(datasets.GeneratorBasedBuilder):
251
  gen_kwargs={
252
  "features_file": data_dir[task_name][split_name]["features_file"],
253
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
254
- "split": split_name,
255
  }
256
  )
257
  )
@@ -263,7 +259,6 @@ class P3(datasets.GeneratorBasedBuilder):
263
  gen_kwargs={
264
  "features_file": data_dir[task_name][split_name]["features_file"],
265
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
266
- "split": split_name,
267
  }
268
  )
269
  )
@@ -276,14 +271,13 @@ class P3(datasets.GeneratorBasedBuilder):
276
  gen_kwargs={
277
  "features_file": data_dir[task_name][special_split_name]["features_file"],
278
  "tfrecord": data_dir[task_name][special_split_name]["tfrecord"],
279
- "split": special_split_name,
280
  }
281
  )
282
  )
283
  return split_generators
284
 
285
 
286
- def _generate_examples(self, features_file, tfrecord, split):
287
  """This function returns the examples in the raw (text) form."""
288
  _FEAT_MAPPING_FUNCTIONS = {
289
  "answer_choices": lambda x: [choice.decode("utf-8") for choice in x],
@@ -297,7 +291,7 @@ class P3(datasets.GeneratorBasedBuilder):
297
  }
298
 
299
  key = 0
300
- ds = load_cached_task(features_file, tfrecord, split)
301
  for ex in ds.as_numpy_iterator():
302
  ex_dict = {}
303
  for feat_name, feat_value in ex.items():
 
78
  # )
79
  # return ds
80
 
81
+ def load_cached_task(features_file, tfrecord):
82
  # # TODO(Victor): this info.*.json is actually done twice... -> factorize
83
  # with tf.io.gfile.GFile(os.path.join(cache_dir, f"info.{split}.json")) as f:
84
  with tf.io.gfile.GFile(features_file) as f:
 
100
  feat: _feature_config(**desc) for feat, desc in features.items()
101
  }
102
 
 
 
 
103
  ds = tf.data.TFRecordDataset(tf.io.gfile.glob([tfrecord]))
104
  ds = ds.map(
105
  lambda pb: tf.io.parse_single_example(pb, feature_description),
 
113
  )
114
  return ds
115
 
116
+
117
  def find_task_splits_and_features():
118
  """Find the available tasks under ./data and their available splits and features."""
119
  task_and_their_splits = defaultdict(dict)
 
237
  gen_kwargs={
238
  "features_file": data_dir[task_name][split_name]["features_file"],
239
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
 
240
  }
241
  )
242
  )
 
248
  gen_kwargs={
249
  "features_file": data_dir[task_name][split_name]["features_file"],
250
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
 
251
  }
252
  )
253
  )
 
259
  gen_kwargs={
260
  "features_file": data_dir[task_name][split_name]["features_file"],
261
  "tfrecord": data_dir[task_name][split_name]["tfrecord"],
 
262
  }
263
  )
264
  )
 
271
  gen_kwargs={
272
  "features_file": data_dir[task_name][special_split_name]["features_file"],
273
  "tfrecord": data_dir[task_name][special_split_name]["tfrecord"],
 
274
  }
275
  )
276
  )
277
  return split_generators
278
 
279
 
280
+ def _generate_examples(self, features_file, tfrecord):
281
  """This function returns the examples in the raw (text) form."""
282
  _FEAT_MAPPING_FUNCTIONS = {
283
  "answer_choices": lambda x: [choice.decode("utf-8") for choice in x],
 
291
  }
292
 
293
  key = 0
294
+ ds = load_cached_task(features_file, tfrecord)
295
  for ex in ds.as_numpy_iterator():
296
  ex_dict = {}
297
  for feat_name, feat_value in ex.items():