Python tensorflow.get_seed() Examples
The following are 18
code examples of tensorflow.get_seed().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
tensorflow
, or try the search function
.
.
Example #1
| Source File: graph_builder.py From object_detection_kitti with Apache License 2.0 | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #2
| Source File: graph_builder.py From multilabel-image-classification-tensorflow with MIT License | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
cache_vectors_locally=False,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #3
| Source File: graph_builder.py From g-tensorflow-models with Apache License 2.0 | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
cache_vectors_locally=False,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #4
| Source File: graph_builder.py From HumanRecognition with MIT License | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #5
| Source File: graph_builder.py From object_detection_with_tensorflow with MIT License | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
cache_vectors_locally=False,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #6
| Source File: graph_builder.py From hands-detection with MIT License | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #7
| Source File: graph_builder.py From Gun-Detector with Apache License 2.0 | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
cache_vectors_locally=False,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #8
| Source File: graph_builder.py From yolo_v2 with Apache License 2.0 | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
cache_vectors_locally=False,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #9
| Source File: graph_builder.py From DOTA_models with Apache License 2.0 | 6 votes |
|
def AddPretrainedEmbeddings(self, index, embeddings_path, task_context):
"""Embeddings at the given index will be set to pretrained values."""
def _Initializer(shape, dtype=tf.float32, partition_info=None):
"""Variable initializer that loads pretrained embeddings."""
unused_dtype = dtype
seed1, seed2 = tf.get_seed(self._seed)
t = gen_parser_ops.word_embedding_initializer(
vectors=embeddings_path,
task_context=task_context,
embedding_init=self._embedding_init,
seed=seed1,
seed2=seed2)
t.set_shape(shape)
return t
self._pretrained_embeddings[index] = _Initializer
Example #10
| Source File: network_units.py From hands-detection with MIT License | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
num_special_embeddings=1,
embedding_init=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(name, initializer=tf.reshape(embeddings, shape))
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed))
Example #11
| Source File: network_units.py From object_detection_kitti with Apache License 2.0 | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
num_special_embeddings=1,
embedding_init=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(name, initializer=tf.reshape(embeddings, shape))
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed))
Example #12
| Source File: network_units.py From DOTA_models with Apache License 2.0 | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = dragnn_ops.dragnn_embedding_initializer(
embedding_input=feature_spec.pretrained_embedding_matrix.part[0]
.file_pattern,
vocab=feature_spec.vocab.part[0].file_pattern,
scaling_coefficient=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(name, initializer=tf.reshape(embeddings, shape))
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed))
Example #13
| Source File: network_units.py From object_detection_with_tensorflow with MIT License | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
num_special_embeddings=1,
embedding_init=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(
name,
initializer=tf.reshape(embeddings, shape),
trainable=not feature_spec.is_constant)
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed),
trainable=not feature_spec.is_constant)
Example #14
| Source File: network_units.py From Gun-Detector with Apache License 2.0 | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
num_special_embeddings=1,
embedding_init=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(
name,
initializer=tf.reshape(embeddings, shape),
trainable=not feature_spec.is_constant)
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed),
trainable=not feature_spec.is_constant)
Example #15
| Source File: network_units.py From HumanRecognition with MIT License | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = dragnn_ops.dragnn_embedding_initializer(
embedding_input=feature_spec.pretrained_embedding_matrix.part[0]
.file_pattern,
vocab=feature_spec.vocab.part[0].file_pattern,
scaling_coefficient=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(name, initializer=tf.reshape(embeddings, shape))
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed))
Example #16
| Source File: network_units.py From g-tensorflow-models with Apache License 2.0 | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
row_num = feature_spec.vocabulary_size + 1
shape = [row_num, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
override_num_embeddings=row_num,
embedding_init=0.0, # zero out rows with no pretrained values
seed=seed1,
seed2=seed2)
return tf.get_variable(
name,
initializer=tf.reshape(embeddings, shape),
trainable=not feature_spec.is_constant)
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed),
trainable=not feature_spec.is_constant)
Example #17
| Source File: network_units.py From yolo_v2 with Apache License 2.0 | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
shape = [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
num_special_embeddings=1,
embedding_init=1.0,
seed=seed1,
seed2=seed2)
return tf.get_variable(
name,
initializer=tf.reshape(embeddings, shape),
trainable=not feature_spec.is_constant)
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed),
trainable=not feature_spec.is_constant)
Example #18
| Source File: network_units.py From multilabel-image-classification-tensorflow with MIT License | 4 votes |
|
def add_embeddings(channel_id, feature_spec, seed=None):
"""Adds a variable for the embedding of a given fixed feature.
Supports pre-trained or randomly initialized embeddings In both cases, extra
vector is reserved for out-of-vocabulary words, so the embedding matrix has
the size of [feature_spec.vocabulary_size + 1, feature_spec.embedding_dim].
Args:
channel_id: Numeric id of the fixed feature channel
feature_spec: Feature spec protobuf of type FixedFeatureChannel
seed: used for random initializer
Returns:
tf.Variable object corresponding to the embedding for that feature.
Raises:
RuntimeError: if more the pretrained embeddings are specified in resources
containing more than one part.
"""
check.Gt(feature_spec.embedding_dim, 0,
'Embeddings requested for non-embedded feature: %s' % feature_spec)
name = fixed_embeddings_name(channel_id)
row_num = feature_spec.vocabulary_size + 1
shape = [row_num, feature_spec.embedding_dim]
if feature_spec.HasField('pretrained_embedding_matrix'):
if len(feature_spec.pretrained_embedding_matrix.part) > 1:
raise RuntimeError('pretrained_embedding_matrix resource contains '
'more than one part:\n%s',
str(feature_spec.pretrained_embedding_matrix))
if len(feature_spec.vocab.part) > 1:
raise RuntimeError('vocab resource contains more than one part:\n%s',
str(feature_spec.vocab))
seed1, seed2 = tf.get_seed(seed)
embeddings = syntaxnet_ops.word_embedding_initializer(
vectors=feature_spec.pretrained_embedding_matrix.part[0].file_pattern,
vocabulary=feature_spec.vocab.part[0].file_pattern,
override_num_embeddings=row_num,
embedding_init=0.0, # zero out rows with no pretrained values
seed=seed1,
seed2=seed2)
return tf.get_variable(
name,
initializer=tf.reshape(embeddings, shape),
trainable=not feature_spec.is_constant)
else:
return tf.get_variable(
name,
shape,
initializer=tf.random_normal_initializer(
stddev=1.0 / feature_spec.embedding_dim**.5, seed=seed),
trainable=not feature_spec.is_constant)
