def get_classification_loss(logits, targets, softmax_loss_function=None):
bucket_outputs = logits
if softmax_loss_function is None:
assert len(bucket_outputs) == len(targets) == 1
# We need to make target an int64-tensor and set its shape.
bucket_target = array_ops.reshape(math_ops.to_int64(targets[0]), [-1])
crossent = nn_ops.sparse_softmax_cross_entropy_with_logits(logits=bucket_outputs[0],
labels=bucket_target)
else:
assert len(bucket_outputs) == len(targets) == 1
crossent = softmax_loss_function(bucket_outputs[0], targets[0])
batch_size = array_ops.shape(targets[0])[0]
loss = tf.reduce_sum(crossent) / math_ops.cast(batch_size, dtypes.float32)
return loss
python类to_int64()的实例源码
def _get_eval_ops(self, features, targets, metrics):
features, spec = data_ops.ParseDataTensorOrDict(features)
labels = data_ops.ParseLabelTensorOrDict(targets)
graph_builder = self.graph_builder_class(
self.params, device_assigner=self.device_assigner, training=False,
**self.construction_args)
probabilities = graph_builder.inference_graph(features, data_spec=spec)
# One-hot the labels.
if not self.params.regression:
labels = math_ops.to_int64(array_ops.one_hot(math_ops.to_int64(
array_ops.squeeze(labels)), self.params.num_classes, 1, 0))
if metrics is None:
metrics = {self.accuracy_metric:
eval_metrics.get_metric(self.accuracy_metric)}
result = {}
for name, metric in six.iteritems(metrics):
result[name] = metric(probabilities, labels)
return result
def _lengths_to_masks(lengths, max_length):
"""Creates a binary matrix that can be used to mask away padding.
Args:
lengths: A vector of integers representing lengths.
max_length: An integer indicating the maximum length. All values in
lengths should be less than max_length.
Returns:
masks: Masks that can be used to get rid of padding.
"""
tiled_ranges = array_ops.tile(
array_ops.expand_dims(math_ops.range(max_length), 0),
[array_ops.shape(lengths)[0], 1])
lengths = array_ops.expand_dims(lengths, 1)
masks = math_ops.to_float(
math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
return masks
def assert_integer_form(
x, data=None, summarize=None, message=None, name="assert_integer_form"):
"""Assert that x has integer components (or floats equal to integers).
Args:
x: Numeric `Tensor`
data: The tensors to print out if the condition is `False`. Defaults to
error message and first few entries of `x` and `y`.
summarize: Print this many entries of each tensor.
message: A string to prefix to the default message.
name: A name for this operation (optional).
Returns:
Op raising `InvalidArgumentError` if round(x) != x.
"""
message = message or "x has non-integer components"
x = ops.convert_to_tensor(x, name="x")
casted_x = math_ops.to_int64(x)
return check_ops.assert_equal(
x, math_ops.cast(math_ops.round(casted_x), x.dtype),
data=data, summarize=summarize, message=message, name=name)
def _lengths_to_masks(lengths, max_length):
"""Creates a binary matrix that can be used to mask away padding.
Args:
lengths: A vector of integers representing lengths.
max_length: An integer indicating the maximum length. All values in
lengths should be less than max_length.
Returns:
masks: Masks that can be used to get rid of padding.
"""
tiled_ranges = array_ops.tile(
array_ops.expand_dims(math_ops.range(max_length), 0),
[array_ops.shape(lengths)[0], 1])
lengths = array_ops.expand_dims(lengths, 1)
masks = math_ops.to_float(
math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
return masks
def assert_integer_form(
x, data=None, summarize=None, message=None, name="assert_integer_form"):
"""Assert that x has integer components (or floats equal to integers).
Args:
x: Numeric `Tensor`
data: The tensors to print out if the condition is `False`. Defaults to
error message and first few entries of `x` and `y`.
summarize: Print this many entries of each tensor.
message: A string to prefix to the default message.
name: A name for this operation (optional).
Returns:
Op raising `InvalidArgumentError` if round(x) != x.
"""
message = message or "x has non-integer components"
x = ops.convert_to_tensor(x, name="x")
casted_x = math_ops.to_int64(x)
return check_ops.assert_equal(
x, math_ops.cast(math_ops.round(casted_x), x.dtype),
data=data, summarize=summarize, message=message, name=name)
def _lengths_to_masks(lengths, max_length):
"""Creates a binary matrix that can be used to mask away padding.
Args:
lengths: A vector of integers representing lengths.
max_length: An integer indicating the maximum length. All values in
lengths should be less than max_length.
Returns:
masks: Masks that can be used to get rid of padding.
"""
tiled_ranges = array_ops.tile(
array_ops.expand_dims(math_ops.range(max_length), 0),
[array_ops.shape(lengths)[0], 1])
lengths = array_ops.expand_dims(lengths, 1)
masks = math_ops.to_float(
math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
return masks
crf.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 21
收藏 0
点赞 0
评论 0
def _lengths_to_masks(lengths, max_length):
"""Creates a binary matrix that can be used to mask away padding.
Args:
lengths: A vector of integers representing lengths.
max_length: An integer indicating the maximum length. All values in
lengths should be less than max_length.
Returns:
masks: Masks that can be used to get rid of padding.
"""
tiled_ranges = array_ops.tile(
array_ops.expand_dims(math_ops.range(max_length), 0),
[array_ops.shape(lengths)[0], 1])
lengths = array_ops.expand_dims(lengths, 1)
masks = math_ops.to_float(
math_ops.to_int64(tiled_ranges) < math_ops.to_int64(lengths))
return masks
distribution_util.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 20
收藏 0
点赞 0
评论 0
def assert_integer_form(
x, data=None, summarize=None, message=None, name="assert_integer_form"):
"""Assert that x has integer components (or floats equal to integers).
Args:
x: Numeric `Tensor`
data: The tensors to print out if the condition is `False`. Defaults to
error message and first few entries of `x` and `y`.
summarize: Print this many entries of each tensor.
message: A string to prefix to the default message.
name: A name for this operation (optional).
Returns:
Op raising `InvalidArgumentError` if round(x) != x.
"""
message = message or "x has non-integer components"
x = ops.convert_to_tensor(x, name="x")
casted_x = math_ops.to_int64(x)
return check_ops.assert_equal(
x, math_ops.cast(math_ops.round(casted_x), x.dtype),
data=data, summarize=summarize, message=message, name=name)
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, depth)
lengths: A tensor of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
input_shape = tensor_shape.matrix(None, None)
for input_ in input_seq:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(input_seq)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r in result:
r.set_shape(input_shape)
return result
def get_sequence_loss(logits, targets, weights, softmax_loss_function=None, per_example_loss=False):
if per_example_loss:
assert len(logits) == len(targets)
# We need to make target and int64-tensor and set its shape.
bucket_target = [array_ops.reshape(math_ops.to_int64(x), [-1]) for x in targets]
crossent = sequence_loss_by_example(logits, bucket_target, weights,
softmax_loss_function=softmax_loss_function)
else:
assert len(logits) == len(targets)
bucket_target = [array_ops.reshape(math_ops.to_int64(x), [-1]) for x in targets]
crossent = sequence_loss_by_batch(logits, bucket_target, weights,
softmax_loss_function=softmax_loss_function)
return crossent
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, depth)
lengths: A tensor of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
input_shape = tensor_shape.matrix(None, None)
for input_ in input_seq:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(input_seq)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r in result:
r.set_shape(input_shape)
return result
def _get_eval_ops(self, features, targets, metrics):
features, _, spec = data_ops.ParseDataTensorOrDict(features)
labels = data_ops.ParseLabelTensorOrDict(targets)
_assert_float32(features)
_assert_float32(labels)
graph_builder = self.graph_builder_class(
self.params, device_assigner=self.device_assigner, training=False,
**self.construction_args)
probabilities = graph_builder.inference_graph(features, data_spec=spec)
# One-hot the labels.
if not self.params.regression:
labels = math_ops.to_int64(array_ops.one_hot(math_ops.to_int64(
array_ops.squeeze(labels)), self.params.num_classes, 1, 0))
if metrics is None:
metrics = {self.accuracy_metric:
eval_metrics.get_metric(self.accuracy_metric)}
result = {}
for name, metric in six.iteritems(metrics):
result[name] = metric(probabilities, labels)
return result
def to_dnn_input_layer(self,
input_tensor,
weight_collections=None,
trainable=True):
return array_ops.reshape(
array_ops.one_hot(
math_ops.to_int64(input_tensor),
self.length,
1.,
0.,
name="one_hot"), [-1, self.length * self.source_column.dimension],
name="reshape")
def _select_class_id(ids, selected_id):
"""Filter all but `selected_id` out of `ids`.
Args:
ids: `int64` `Tensor` or `SparseTensor` of IDs.
selected_id: Int id to select.
Returns:
`SparseTensor` of same dimensions as `ids`, except for the last dimension,
which might be smaller. This contains only the entries equal to
`selected_id`.
"""
if isinstance(ids, (ops.SparseTensor, ops.SparseTensorValue)):
return sparse_ops.sparse_retain(
ids, math_ops.equal(ids.values, selected_id))
# TODO(ptucker): Make this more efficient, maybe add a sparse version of
# tf.equal and tf.reduce_any?
# Shape of filled IDs is the same as `ids` with the last dim collapsed to 1.
ids_shape = array_ops.shape(ids)
ids_last_dim = array_ops.size(ids_shape) - 1
filled_selected_id_shape = math_ops.reduced_shape(
ids_shape, array_ops.reshape(ids_last_dim, [1]))
# Intersect `ids` with the selected ID.
filled_selected_id = array_ops.fill(
filled_selected_id_shape, math_ops.to_int64(selected_id))
return set_ops.set_intersection(filled_selected_id, ids)
def _logits_to_predictions(self, logits):
"""See `_MultiClassHead`."""
predictions = {prediction_key.PredictionKey.LOGITS: logits}
if self.logits_dimension == 1:
predictions[prediction_key.PredictionKey.LOGISTIC] = math_ops.sigmoid(
logits)
logits = array_ops.concat(1, [array_ops.zeros_like(logits), logits])
predictions[prediction_key.PredictionKey.PROBABILITIES] = math_ops.sigmoid(
logits)
predictions[prediction_key.PredictionKey.CLASSES] = math_ops.to_int64(
math_ops.greater(logits, 0))
return predictions
def _to_dnn_input_layer(self,
input_tensor,
weight_collections=None,
trainable=True,
output_rank=2):
if output_rank != 2:
raise ValueError("BucketizedColumn currently only supports output_rank=2")
return array_ops.reshape(
array_ops.one_hot(
math_ops.to_int64(input_tensor),
self.length,
1.,
0.,
name="one_hot"), [-1, self.length * self.source_column.dimension],
name="reshape")
def _select_class_id(ids, selected_id):
"""Filter all but `selected_id` out of `ids`.
Args:
ids: `int64` `Tensor` or `SparseTensor` of IDs.
selected_id: Int id to select.
Returns:
`SparseTensor` of same dimensions as `ids`. This contains only the entries
equal to `selected_id`.
"""
if isinstance(
ids, (sparse_tensor.SparseTensor, sparse_tensor.SparseTensorValue)):
return sparse_ops.sparse_retain(
ids, math_ops.equal(ids.values, selected_id))
# TODO(ptucker): Make this more efficient, maybe add a sparse version of
# tf.equal and tf.reduce_any?
# Shape of filled IDs is the same as `ids` with the last dim collapsed to 1.
ids_shape = array_ops.shape(ids, out_type=dtypes.int64)
ids_last_dim = array_ops.size(ids_shape) - 1
filled_selected_id_shape = math_ops.reduced_shape(
ids_shape, array_ops.reshape(ids_last_dim, [1]))
# Intersect `ids` with the selected ID.
filled_selected_id = array_ops.fill(
filled_selected_id_shape, math_ops.to_int64(selected_id))
result = set_ops.set_intersection(filled_selected_id, ids)
return sparse_tensor.SparseTensor(
indices=result.indices, values=result.values, shape=ids_shape)
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, depth)
lengths: A tensor of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
input_shape = tensor_shape.matrix(None, None)
for input_ in input_seq:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(input_seq)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r in result:
r.set_shape(input_shape)
return result
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, depth)
lengths: A tensor of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
input_shape = tensor_shape.matrix(None, None)
for input_ in input_seq:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(input_seq)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r in result:
r.set_shape(input_shape)
return result
def _reverse_seq(input_seq, lengths):
"""Reverse a list of Tensors up to specified lengths.
Args:
input_seq: Sequence of seq_len tensors of dimension (batch_size, n_features)
or nested tuples of tensors.
lengths: A `Tensor` of dimension batch_size, containing lengths for each
sequence in the batch. If "None" is specified, simply reverses
the list.
Returns:
time-reversed sequence
"""
if lengths is None:
return list(reversed(input_seq))
flat_input_seq = tuple(nest.flatten(input_) for input_ in input_seq)
flat_results = [[] for _ in range(len(input_seq))]
for sequence in zip(*flat_input_seq):
input_shape = tensor_shape.unknown_shape(
ndims=sequence[0].get_shape().ndims)
for input_ in sequence:
input_shape.merge_with(input_.get_shape())
input_.set_shape(input_shape)
# Join into (time, batch_size, depth)
s_joined = array_ops.pack(sequence)
# TODO(schuster, ebrevdo): Remove cast when reverse_sequence takes int32
if lengths is not None:
lengths = math_ops.to_int64(lengths)
# Reverse along dimension 0
s_reversed = array_ops.reverse_sequence(s_joined, lengths, 0, 1)
# Split again into list
result = array_ops.unpack(s_reversed)
for r, flat_result in zip(result, flat_results):
r.set_shape(input_shape)
flat_result.append(r)
results = [nest.pack_sequence_as(structure=input_, flat_sequence=flat_result)
for input_, flat_result in zip(input_seq, flat_results)]
return results
def get_classification_loss(logits, targets, softmax_loss_function=None):
bucket_outputs = logits
if softmax_loss_function is None:
assert len(bucket_outputs) == len(targets) == 1
# We need to make target an int64-tensor and set its shape.
bucket_target = array_ops.reshape(math_ops.to_int64(targets[0]), [-1])
crossent = nn_ops.sparse_softmax_cross_entropy_with_logits(bucket_outputs[0], bucket_target)
else:
assert len(bucket_outputs) == len(targets) == 1
crossent = softmax_loss_function(bucket_outputs[0], targets[0])
batch_size = array_ops.shape(targets[0])[0]
loss = tf.reduce_sum(crossent) / math_ops.cast(batch_size, dtypes.float32)
return loss
head.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 21
收藏 0
点赞 0
评论 0
def _logits_to_predictions(self, logits):
"""See `_MultiClassHead`."""
with ops.name_scope(None, "predictions", (logits,)):
return {
prediction_key.PredictionKey.LOGITS:
logits,
prediction_key.PredictionKey.PROBABILITIES:
math_ops.sigmoid(
logits, name=prediction_key.PredictionKey.PROBABILITIES),
prediction_key.PredictionKey.CLASSES:
math_ops.to_int64(
math_ops.greater(logits, 0),
name=prediction_key.PredictionKey.CLASSES)
}
estimator_test.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 23
收藏 0
点赞 0
评论 0
def _build_estimator_for_export_tests(tmpdir):
def _input_fn():
iris = base.load_iris()
return {
'feature': constant_op.constant(
iris.data, dtype=dtypes.float32)
}, constant_op.constant(
iris.target, shape=[150], dtype=dtypes.int32)
feature_columns = [
feature_column_lib.real_valued_column(
'feature', dimension=4)
]
est = linear.LinearRegressor(feature_columns)
est.fit(input_fn=_input_fn, steps=20)
feature_spec = feature_column_lib.create_feature_spec_for_parsing(
feature_columns)
serving_input_fn = input_fn_utils.build_parsing_serving_input_fn(feature_spec)
# hack in an op that uses an asset, in order to test asset export.
# this is not actually valid, of course.
def serving_input_fn_with_asset():
features, labels, inputs = serving_input_fn()
vocab_file_name = os.path.join(tmpdir, 'my_vocab_file')
vocab_file = gfile.GFile(vocab_file_name, mode='w')
vocab_file.write(VOCAB_FILE_CONTENT)
vocab_file.close()
hashtable = lookup.HashTable(
lookup.TextFileStringTableInitializer(vocab_file_name), 'x')
features['bogus_lookup'] = hashtable.lookup(
math_ops.to_int64(features['feature']))
return input_fn_utils.InputFnOps(features, labels, inputs)
return est, serving_input_fn_with_asset
feature_column.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 24
收藏 0
点赞 0
评论 0
def _to_dnn_input_layer(self,
input_tensor,
weight_collections=None,
trainable=True,
output_rank=2):
if output_rank != 2:
raise ValueError("BucketizedColumn currently only supports output_rank=2")
return array_ops.reshape(
array_ops.one_hot(
math_ops.to_int64(input_tensor),
self.length,
1.,
0.,
name="one_hot"), [-1, self.length * self.source_column.dimension],
name="reshape")
feature_column.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 33
收藏 0
点赞 0
评论 0
def to_sparse_tensor(self, input_tensor):
"""Creates a SparseTensor from the bucketized Tensor."""
dimension = self.source_column.dimension
batch_size = array_ops.shape(input_tensor, name="shape")[0]
if dimension > 1:
i1 = array_ops.reshape(
array_ops.tile(
array_ops.expand_dims(
math_ops.range(0, batch_size), 1, name="expand_dims"),
[1, dimension],
name="tile"), [-1],
name="rehsape")
i2 = array_ops.tile(
math_ops.range(0, dimension), [batch_size], name="tile")
# Flatten the bucket indices and unique them across dimensions
# E.g. 2nd dimension indices will range from k to 2*k-1 with k buckets
bucket_indices = array_ops.reshape(
input_tensor, [-1], name="reshape") + self.length * i2
else:
# Simpler indices when dimension=1
i1 = math_ops.range(0, batch_size)
i2 = array_ops.zeros([batch_size], dtype=dtypes.int32, name="zeros")
bucket_indices = array_ops.reshape(input_tensor, [-1], name="reshape")
indices = math_ops.to_int64(array_ops.transpose(array_ops.stack((i1, i2))))
shape = math_ops.to_int64(array_ops.stack([batch_size, dimension]))
sparse_id_values = sparse_tensor_py.SparseTensor(
indices, bucket_indices, shape)
return sparse_id_values
gmm_ops.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 24
收藏 0
点赞 0
评论 0
def _define_distance_to_clusters(self, data):
"""Defines the Mahalanobis distance to the assigned Gaussian."""
# TODO(xavigonzalvo): reuse (input - mean) * cov^-1 * (input -
# mean) from log probability function.
self._all_scores = []
for shard in data:
all_scores = []
shard = array_ops.expand_dims(shard, 0)
for c in xrange(self._num_classes):
if self._covariance_type == FULL_COVARIANCE:
cov = self._covs[c, :, :]
elif self._covariance_type == DIAG_COVARIANCE:
cov = array_ops.diag(self._covs[c, :])
inverse = linalg_ops.matrix_inverse(cov + self._min_var)
inv_cov = array_ops.tile(
array_ops.expand_dims(inverse, 0),
array_ops.stack([self._num_examples, 1, 1]))
diff = array_ops.transpose(shard - self._means[c, :, :], perm=[1, 0, 2])
m_left = math_ops.matmul(diff, inv_cov)
all_scores.append(
math_ops.sqrt(
math_ops.matmul(
m_left, array_ops.transpose(
diff, perm=[0, 2, 1]))))
self._all_scores.append(
array_ops.reshape(
array_ops.concat(all_scores, 1),
array_ops.stack([self._num_examples, self._num_classes])))
# Distance to the associated class.
self._all_scores = array_ops.concat(self._all_scores, 0)
assignments = array_ops.concat(self.assignments(), 0)
rows = math_ops.to_int64(math_ops.range(0, self._num_examples))
indices = array_ops.concat(
[array_ops.expand_dims(rows, 1), array_ops.expand_dims(assignments, 1)],
1)
self._scores = array_ops.gather_nd(self._all_scores, indices)
lstm1d.py 文件源码
项目:DeepLearning_VirtualReality_BigData_Project
作者: rashmitripathi
项目源码
文件源码
阅读 21
收藏 0
点赞 0
评论 0
def ndlstm_base_dynamic(inputs, noutput, scope=None, reverse=False):
"""Run an LSTM, either forward or backward.
This is a 1D LSTM implementation using dynamic_rnn and
the TensorFlow LSTM op.
Args:
inputs: input sequence (length, batch_size, ninput)
noutput: depth of output
scope: optional scope name
reverse: run LSTM in reverse
Returns:
Output sequence (length, batch_size, noutput)
"""
with variable_scope.variable_scope(scope, "SeqLstm", [inputs]):
# TODO(tmb) make batch size, sequence_length dynamic
# example: sequence_length = tf.shape(inputs)[0]
_, batch_size, _ = _shape(inputs)
lstm_cell = core_rnn_cell_impl.BasicLSTMCell(noutput, state_is_tuple=False)
state = array_ops.zeros([batch_size, lstm_cell.state_size])
sequence_length = int(inputs.get_shape()[0])
sequence_lengths = math_ops.to_int64(
array_ops.fill([batch_size], sequence_length))
if reverse:
inputs = array_ops.reverse_v2(inputs, [0])
outputs, _ = rnn.dynamic_rnn(
lstm_cell, inputs, sequence_lengths, state, time_major=True)
if reverse:
outputs = array_ops.reverse_v2(outputs, [0])
return outputs
def sequence_loss_per_sample(logits,
targets,
weights):
"""TODO(nh2tran): docstring.
Weighted cross-entropy loss for a sequence of logits (per example).
Args:
logits: List of 2D Tensors of shape [batch_size x num_decoder_symbols].
targets: List of 1D batch-sized int32 Tensors of the same length as logits.
weights: List of 1D batch-sized float-Tensors of the same length as logits.
average_across_timesteps: If set, divide the returned cost by the total
label weight.
softmax_loss_function: Function (inputs-batch, labels-batch) -> loss-batch
to be used instead of the standard softmax (the default if this is None).
name: Optional name for this operation, default: "sequence_loss_by_example".
Returns:
1D batch-sized float Tensor: The log-perplexity for each sequence.
Raises:
ValueError: If len(logits) is different from len(targets) or len(weights).
"""
#~ with tf.name_scope(name="sequence_loss_by_example",
#~ values=logits + targets + weights):
with ops.op_scope(logits + targets + weights,
None,
"sequence_loss_by_example"):
log_perp_list = []
for logit, target, weight in zip(logits, targets, weights):
target = array_ops.reshape(math_ops.to_int64(target), [-1])
crossent = nn_ops.sparse_softmax_cross_entropy_with_logits(logits=logit,
labels=target)
log_perp_list.append(crossent * weight)
log_perps = math_ops.add_n(log_perp_list)
# average_across_timesteps:
total_size = math_ops.add_n(weights)
total_size += 1e-12 # Just to avoid division by 0 for all-0 weights.
log_perps /= total_size
return log_perps
def generate_task_output(encoder_outputs, additional_inputs, encoder_state, targets,sequence_length, num_decoder_symbols, weights,
buckets, softmax_loss_function=None,
per_example_loss=False, name=None, use_attention=False, scope=None, DNN_at_output=False,
intent_results=None,
tagging_results=None,
train_with_true_label=True,
use_local_context=False,
forward_only=False):
if len(targets) < buckets[-1][1]:
raise ValueError("Length of targets (%d) must be at least that of last"
"bucket (%d)." % (len(targets), buckets[-1][1]))
all_inputs = encoder_outputs + targets + weights
with ops.op_scope(all_inputs, name, "model_with_buckets"):
if scope == 'intent':
logits, regularizers, sampled_intents = intent_results
sampled_tags = list()
elif scope == 'tagging':
logits, regularizers, sampled_tags = tagging_results
sampled_intents = list()
elif scope == 'lm':
with variable_scope.variable_scope(scope + "_generate_sequence_output", reuse=None):
task_inputs = []
if use_local_context:
print ('lm task: use sampled_tag_intent_emb as local context')
task_inputs = [array_ops.concat(1, [additional_input, encoder_output]) for additional_input, encoder_output in zip(additional_inputs, encoder_outputs)]
else:
task_inputs = encoder_outputs
logits, _, regularizers = generate_sequence_output(task_inputs,
encoder_state,
num_decoder_symbols,
sequence_length,
use_attention=use_attention,
DNN_at_output=DNN_at_output,
forward_only=forward_only)
sampled_tags = list()
sampled_intents = list()
if per_example_loss is None:
assert len(logits) == len(targets)
# We need to make target and int64-tensor and set its shape.
bucket_target = [array_ops.reshape(math_ops.to_int64(x), [-1]) for x in targets]
crossent = sequence_loss_by_example(
logits, bucket_target, weights,
softmax_loss_function=softmax_loss_function)
else:
assert len(logits) == len(targets)
bucket_target = [array_ops.reshape(math_ops.to_int64(x), [-1]) for x in targets]
crossent = sequence_loss(
logits, bucket_target, weights,
softmax_loss_function=softmax_loss_function)
crossent_with_regularizers = crossent + 1e-4 * regularizers
return logits, sampled_tags, sampled_intents, crossent_with_regularizers, crossent