Datasets:

Tasks:
Other
Modalities:
Text
Formats:
parquet
Languages:
English
ArXiv:
Tags:
License:
VictorSanh commited on
Commit
856a31b
1 Parent(s): 210a627

breaking down download of files

Browse files
Files changed (1) hide show
  1. P3.py +69 -18
P3.py CHANGED
@@ -41,11 +41,50 @@ _HOMEPAGE = "https://github.com/bigscience-workshop/promptsource"
41
  _DATA_PATH = "data"
42
 
43
 
44
- def load_cached_task(cache_dir, split):
45
- # TODO(Victor): this info.*.json is actually done twice... -> factorize
46
- with tf.io.gfile.GFile(os.path.join(cache_dir, f"info.{split}.json")) as f:
47
- split_info = json.load(f)
48
- features = split_info["features"]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
 
50
  # Use `FixedLenSequenceFeature` for sequences with variable length.
51
  def _feature_config(shape, dtype):
@@ -62,10 +101,10 @@ def load_cached_task(cache_dir, split):
62
  feat: _feature_config(**desc) for feat, desc in features.items()
63
  }
64
 
65
- tfrecords = os.path.join(
66
- cache_dir, f"{split}.tfrecord-*-of-*{split_info['num_shards']}"
67
- )
68
- ds = tf.data.TFRecordDataset(tf.io.gfile.glob(tfrecords))
69
  ds = ds.map(
70
  lambda pb: tf.io.parse_single_example(pb, feature_description),
71
  num_parallel_calls=tf.data.experimental.AUTOTUNE
@@ -78,7 +117,6 @@ def load_cached_task(cache_dir, split):
78
  )
79
  return ds
80
 
81
-
82
  def find_task_splits_and_features():
83
  """Find the available tasks under ./data and their available splits and features."""
84
  task_and_their_splits = defaultdict(dict)
@@ -100,6 +138,7 @@ def find_task_splits_and_features():
100
  with open(os.path.join(folder_path, f"info.{split_name}.json")) as f:
101
  split_info = json.load(f)
102
  features = split_info["features"]
 
103
 
104
  # All splits under the same task have the same features dictionary (and thus the same features list)
105
  if task_and_their_splits[task_name] == {}:
@@ -118,7 +157,16 @@ def find_task_splits_and_features():
118
 
119
 
120
  _TASK_SPLITS_AND_FEATURES = find_task_splits_and_features()
121
- _URLs = {task_name: f"{_DATA_PATH}/{task_name}" for task_name in _TASK_SPLITS_AND_FEATURES.keys()}
 
 
 
 
 
 
 
 
 
122
 
123
 
124
  class P3Config(datasets.BuilderConfig):
@@ -184,13 +232,13 @@ class P3(datasets.GeneratorBasedBuilder):
184
  def _split_generators(self, dl_manager):
185
  split_generators = []
186
  data_dir = dl_manager.download_and_extract(_URLs)
187
- import pdb; pdb.set_trace()
188
  if "train" in self.config.splits:
189
  split_generators.append(
190
  datasets.SplitGenerator(
191
  name=datasets.Split.TRAIN,
192
  gen_kwargs={
193
- "data_folder": data_dir,
 
194
  "split": "train",
195
  }
196
  )
@@ -200,7 +248,8 @@ class P3(datasets.GeneratorBasedBuilder):
200
  datasets.SplitGenerator(
201
  name=datasets.Split.VALIDATION,
202
  gen_kwargs={
203
- "data_folder": data_dir,
 
204
  "split": "validation",
205
  }
206
  )
@@ -210,7 +259,8 @@ class P3(datasets.GeneratorBasedBuilder):
210
  datasets.SplitGenerator(
211
  name=datasets.Split.TEST,
212
  gen_kwargs={
213
- "data_folder": data_dir,
 
214
  "split": "test",
215
  }
216
  )
@@ -222,7 +272,8 @@ class P3(datasets.GeneratorBasedBuilder):
222
  datasets.SplitGenerator(
223
  name=datasets.Split(special_split_name),
224
  gen_kwargs={
225
- "data_folder": data_dir,
 
226
  "split": special_split_name,
227
  }
228
  )
@@ -230,7 +281,7 @@ class P3(datasets.GeneratorBasedBuilder):
230
  return split_generators
231
 
232
 
233
- def _generate_examples(self, data_folder, split):
234
  """This function returns the examples in the raw (text) form."""
235
  _FEAT_MAPPING_FUNCTIONS = {
236
  "answer_choices": lambda x: [choice.decode("utf-8") for choice in x],
@@ -244,7 +295,7 @@ class P3(datasets.GeneratorBasedBuilder):
244
  }
245
 
246
  key = 0
247
- ds = load_cached_task(data_folder, split)
248
  for ex in ds.as_numpy_iterator():
249
  ex_dict = {}
250
  for feat_name, feat_value in ex.items():
 
41
  _DATA_PATH = "data"
42
 
43
 
44
+ # def load_cached_task(cache_dir, split):
45
+ # # TODO(Victor): this info.*.json is actually done twice... -> factorize
46
+ # with tf.io.gfile.GFile(os.path.join(cache_dir, f"info.{split}.json")) as f:
47
+ # split_info = json.load(f)
48
+ # features = split_info["features"]
49
+
50
+ # # Use `FixedLenSequenceFeature` for sequences with variable length.
51
+ # def _feature_config(shape, dtype):
52
+ # if dtype in ("int32", "bool"):
53
+ # # int32 and bool are stored as int64 in the tf.train.Example protobuf.
54
+ # dtype = "int64"
55
+ # if shape and shape[0] is None:
56
+ # return tf.io.FixedLenSequenceFeature(
57
+ # shape[1:], dtype, allow_missing=True
58
+ # )
59
+ # return tf.io.FixedLenFeature(shape, dtype)
60
+
61
+ # feature_description = {
62
+ # feat: _feature_config(**desc) for feat, desc in features.items()
63
+ # }
64
+
65
+ # tfrecords = os.path.join(
66
+ # cache_dir, f"{split}.tfrecord-*-of-*{split_info['num_shards']}"
67
+ # )
68
+ # ds = tf.data.TFRecordDataset(tf.io.gfile.glob(tfrecords))
69
+ # ds = ds.map(
70
+ # lambda pb: tf.io.parse_single_example(pb, feature_description),
71
+ # num_parallel_calls=tf.data.experimental.AUTOTUNE
72
+ # )
73
+ # # Cast features back to the types from the info JSON since some features
74
+ # # must be cast for storage (e.g., in32 is stored as int64).
75
+ # ds = ds.map(
76
+ # lambda x: {k: tf.cast(v, features[k]["dtype"]) for k, v in x.items()},
77
+ # num_parallel_calls=tf.data.experimental.AUTOTUNE
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
+ # split_info = json.load(f)
85
+ # features = split_info["features"]
86
+ with tf.io.gfile.GFile(features_file) as f:
87
+ features = json.load(f)
88
 
89
  # Use `FixedLenSequenceFeature` for sequences with variable length.
90
  def _feature_config(shape, dtype):
 
101
  feat: _feature_config(**desc) for feat, desc in features.items()
102
  }
103
 
104
+ # tfrecords = os.path.join(
105
+ # cache_dir, f"{split}.tfrecord-*-of-*{split_info['num_shards']}"
106
+ # )
107
+ ds = tf.data.TFRecordDataset(tf.io.gfile.glob([tfrecord]))
108
  ds = ds.map(
109
  lambda pb: tf.io.parse_single_example(pb, feature_description),
110
  num_parallel_calls=tf.data.experimental.AUTOTUNE
 
117
  )
118
  return ds
119
 
 
120
  def find_task_splits_and_features():
121
  """Find the available tasks under ./data and their available splits and features."""
122
  task_and_their_splits = defaultdict(dict)
 
138
  with open(os.path.join(folder_path, f"info.{split_name}.json")) as f:
139
  split_info = json.load(f)
140
  features = split_info["features"]
141
+ assert split_info["num_shards"] == 1
142
 
143
  # All splits under the same task have the same features dictionary (and thus the same features list)
144
  if task_and_their_splits[task_name] == {}:
 
157
 
158
 
159
  _TASK_SPLITS_AND_FEATURES = find_task_splits_and_features()
160
+ _URLs = {
161
+ task_name: {
162
+ split_name: {
163
+ "tfrecord": f"{_DATA_PATH}/{task_name}/{split_name}.tfrecord-00000-of-00001",
164
+ "features_file": f"{_DATA_PATH}/{task_name}/info.{split_name}.json",
165
+ }
166
+ for split_name in splits_and_features["splits"]
167
+ }
168
+ for task_name, splits_and_features in _TASK_SPLITS_AND_FEATURES.items()
169
+ }
170
 
171
 
172
  class P3Config(datasets.BuilderConfig):
 
232
  def _split_generators(self, dl_manager):
233
  split_generators = []
234
  data_dir = dl_manager.download_and_extract(_URLs)
 
235
  if "train" in self.config.splits:
236
  split_generators.append(
237
  datasets.SplitGenerator(
238
  name=datasets.Split.TRAIN,
239
  gen_kwargs={
240
+ "features_file": data_dir["features_file"],
241
+ "tfrecord": data_dir["tfrecord"],
242
  "split": "train",
243
  }
244
  )
 
248
  datasets.SplitGenerator(
249
  name=datasets.Split.VALIDATION,
250
  gen_kwargs={
251
+ "features_file": data_dir["features_file"],
252
+ "tfrecord": data_dir["tfrecord"],
253
  "split": "validation",
254
  }
255
  )
 
259
  datasets.SplitGenerator(
260
  name=datasets.Split.TEST,
261
  gen_kwargs={
262
+ "features_file": data_dir["features_file"],
263
+ "tfrecord": data_dir["tfrecord"],
264
  "split": "test",
265
  }
266
  )
 
272
  datasets.SplitGenerator(
273
  name=datasets.Split(special_split_name),
274
  gen_kwargs={
275
+ "features_file": data_dir["features_file"],
276
+ "tfrecord": data_dir["tfrecord"],
277
  "split": special_split_name,
278
  }
279
  )
 
281
  return split_generators
282
 
283
 
284
+ def _generate_examples(self, features_file, tfrecord, split):
285
  """This function returns the examples in the raw (text) form."""
286
  _FEAT_MAPPING_FUNCTIONS = {
287
  "answer_choices": lambda x: [choice.decode("utf-8") for choice in x],
 
295
  }
296
 
297
  key = 0
298
+ ds = load_cached_task(features_file, tfrecord, split)
299
  for ex in ds.as_numpy_iterator():
300
  ex_dict = {}
301
  for feat_name, feat_value in ex.items():