James Briggs commited on
Commit
66d4de0
1 Parent(s): e554e66

Changed data script so it takes a list of files instead of a directory

Browse files
Files changed (1) hide show
  1. wikipedia-bert-128.py +4 -6
wikipedia-bert-128.py CHANGED
@@ -117,18 +117,17 @@ class WikipediaBERT128(datasets.GeneratorBasedBuilder):
117
  # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLs
118
  # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
119
  # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
120
- data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
121
  return [
122
  datasets.SplitGenerator(
123
  name=datasets.Split.TRAIN,
124
  # These kwargs will be passed to _generate_examples
125
  gen_kwargs={
126
- "filepath": data_dir,
127
  },
128
  ),
129
  ]
130
 
131
- def _generate_examples(self, filepath):
132
  """ Yields examples as (key, example) tuples. """
133
  # This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
134
  # The `key` is here for legacy reason (tfds) and is not important in itself.
@@ -142,10 +141,8 @@ class WikipediaBERT128(datasets.GeneratorBasedBuilder):
142
  'masked_lm_ids', # masked_lm_labels=None : label of masked tokens with padding as 0.
143
  'next_sentence_labels' # next_sentence_label=None : 1 if next sentence, 0 otherwise
144
  )
145
- tfrecords = Path(filepath).glob("*.tfrecord")
146
-
147
  highest_id_ = -1
148
- for rec in tfrecords:
149
  reader = tfrecord_loader(rec, None, list(TFRECORD_KEYS))
150
  for id_, d in enumerate(reader, start=highest_id_+1):
151
  highest_id_ = id_
@@ -164,3 +161,4 @@ class WikipediaBERT128(datasets.GeneratorBasedBuilder):
164
  "labels": labels,
165
  "next_sentence_label": d["next_sentence_labels"]
166
  }
 
 
117
  # dl_manager is a datasets.download.DownloadManager that can be used to download and extract URLs
118
  # It can accept any type or nested list/dict and will give back the same structure with the url replaced with path to local files.
119
  # By default the archives will be extracted and a path to a cached folder where they are extracted is returned instead of the archive
 
120
  return [
121
  datasets.SplitGenerator(
122
  name=datasets.Split.TRAIN,
123
  # These kwargs will be passed to _generate_examples
124
  gen_kwargs={
125
+ "data_files": self.config.data_files["train"],
126
  },
127
  ),
128
  ]
129
 
130
+ def _generate_examples(self, data_files):
131
  """ Yields examples as (key, example) tuples. """
132
  # This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
133
  # The `key` is here for legacy reason (tfds) and is not important in itself.
 
141
  'masked_lm_ids', # masked_lm_labels=None : label of masked tokens with padding as 0.
142
  'next_sentence_labels' # next_sentence_label=None : 1 if next sentence, 0 otherwise
143
  )
 
 
144
  highest_id_ = -1
145
+ for rec in data_files:
146
  reader = tfrecord_loader(rec, None, list(TFRECORD_KEYS))
147
  for id_, d in enumerate(reader, start=highest_id_+1):
148
  highest_id_ = id_
 
161
  "labels": labels,
162
  "next_sentence_label": d["next_sentence_labels"]
163
  }
164
+