def prepare_datasets(dataset_name=gin.REQUIRED,
shuffle_input_sentences=False,
num_eval_examples=2000,
batch_size=32):
"""Create batched, properly-formatted datasets from the TFDS datasets.
Args:
dataset_name: Name of TFDS dataset.
shuffle_input_sentences: Not used during evaluation, but arg still needed
for gin compatibility.
num_eval_examples: Number of examples to use during evaluation. For the
nolabel evaluation, this is also the number of distractors we choose
between.
batch_size: Batch size.
Returns:
A dictionary mapping from the dataset split to a Dataset object.
"""
del shuffle_input_sentences
splits_to_load = {
'valid_nolabel': 'train[:2%]',
'train_nolabel': 'train[2%:4%]',
'valid2018': rocstories_sentence_embeddings.VALIDATION_2018,
'valid2016': rocstories_sentence_embeddings.VALIDATION_2016
}
datasets = tfds.load(dataset_name,
data_dir=FLAGS.data_dir,
split=splits_to_load,
download=False)
emb_matrices = {}
valid_nolabel_ds = utils.build_train_style_dataset(
datasets['valid_nolabel'],
batch_size,
False,
num_examples=num_eval_examples,
is_training=False)
datasets['valid_nolabel'], emb_matrices['valid_nolabel'] = valid_nolabel_ds
train_nolabel_ds = utils.build_train_style_dataset(
datasets['train_nolabel'],
batch_size,
False,
num_examples=num_eval_examples,
is_training=False)
datasets['train_nolabel'], emb_matrices['train_nolabel'] = train_nolabel_ds
# Convert official evaluation datasets to validation data format. There are no
# embedding matrices involved here since the task has only two possible next
# sentences to pick between for each example. Ignore num_eval_examples and use
# the full datasets for these.
datasets['valid2018'] = utils.build_validation_dataset(
datasets['valid2018'])
datasets['valid2016'] = utils.build_validation_dataset(
datasets['valid2016'])
return datasets, emb_matrices
def prepare_dataset(dataset_name=gin.REQUIRED,
shuffle_input_sentences=False,
num_eval_examples=2000,
batch_size=32):
"""Create batched, properly-formatted datasets from the TFDS datasets.
Args:
dataset_name: Name of TFDS dataset.
shuffle_input_sentences: Not used during evaluation, but arg still needed
for gin compatibility.
num_eval_examples: Number of examples to use during evaluation. For the
nolabel evaluation, this is also the number of distractors we choose
between.
batch_size: Batch size.
Returns:
The validation dataset, the story identifiers for each story in the
embedding matrix, and the embedding matrix.
"""
del num_eval_examples
del shuffle_input_sentences
splits_to_load = [
tfds.Split.TRAIN,
rocstories_sentence_embeddings.VALIDATION_2018,
]
tfds_train, tfds_valid = tfds.load(dataset_name,
data_dir=FLAGS.data_dir,
split=splits_to_load)
_, train_embs, train_story_ids = utils.build_train_style_dataset(
tfds_train,
batch_size,
shuffle_input_sentences=False,
return_ids=True,
is_training=False)
out = build_all_distractor_valid_dataset(tfds_valid, batch_size=batch_size)
valid_dataset, valid_embs, valid_story_ids = out
all_story_ids = valid_story_ids + train_story_ids
all_emb_matrix = tf.concat([valid_embs, train_embs], axis=0)
return valid_dataset, all_story_ids, all_emb_matrix
def prepare_datasets(dataset_name=gin.REQUIRED,
shuffle_input_sentences=False,
num_eval_examples=2000,
batch_size=32):
"""Create batched, properly-formatted datasets from the TFDS datasets.
Args:
dataset_name: Name of TFDS dataset.
shuffle_input_sentences: If True, the order of the input sentences is
randomized.
num_eval_examples: Number of examples to use during evaluation. For the
nolabel evaluation, this is also the number of distractors we choose
between.
batch_size: Batch size.
Returns:
A dictionary mapping from the dataset split to a Dataset object.
"""
splits_to_load = {
'valid_nolabel': 'train[:2%]',
'train': 'train[2%:]',
'train_nolabel': 'train[2%:4%]',
'valid2018': rocstories_sentence_embeddings.VALIDATION_2018,
'valid2016': rocstories_sentence_embeddings.VALIDATION_2016}
datasets = tfds.load(
dataset_name,
data_dir=FLAGS.data_dir,
split=splits_to_load,
download=False)
emb_matrices = {}
# Convert datasets to expected training data format, and build of the
# embedding matrices.
train_ds = utils.build_train_style_dataset(
datasets['train'], batch_size, shuffle_input_sentences)
datasets['train'], emb_matrices['train'] = train_ds
valid_nolabel_ds = utils.build_train_style_dataset(
datasets['valid_nolabel'], batch_size, False,
num_examples=num_eval_examples)
datasets['valid_nolabel'], emb_matrices['valid_nolabel'] = valid_nolabel_ds
train_nolabel_ds = utils.build_train_style_dataset(
datasets['train_nolabel'], batch_size, False,
num_examples=num_eval_examples)
datasets['train_nolabel'], emb_matrices['train_nolabel'] = train_nolabel_ds
# Convert official evaluation datasets to validation data format. There are no
# embedding matrices involved here since the task has only two possible next
# sentences to pick between for each example.
datasets['valid2018'] = utils.build_validation_dataset(
datasets['valid2018']).take(num_eval_examples)
datasets['valid2016'] = utils.build_validation_dataset(
datasets['valid2016']).take(num_eval_examples)
logging.info('EMBEDDING MATRICES CREATED:')
for key in emb_matrices:
logging.info('%s: %s', key, emb_matrices[key].shape)
return datasets, emb_matrices