1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
|
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Utility functions for training."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.eager import context
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import graph_io
from tensorflow.python.framework import ops
from tensorflow.python.ops import init_ops
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.ops import state_ops
from tensorflow.python.ops import variable_scope
from tensorflow.python.ops import variables
from tensorflow.python.platform import tf_logging as logging
from tensorflow.python.util.tf_export import tf_export
# Picked a long key value to minimize the chance of collision with user defined
# collection keys.
GLOBAL_STEP_READ_KEY = 'global_step_read_op_cache'
# TODO(drpng): remove this after legacy uses are resolved.
write_graph = graph_io.write_graph
@tf_export('train.global_step')
def global_step(sess, global_step_tensor):
"""Small helper to get the global step.
```python
# Create a variable to hold the global_step.
global_step_tensor = tf.Variable(10, trainable=False, name='global_step')
# Create a session.
sess = tf.Session()
# Initialize the variable
sess.run(global_step_tensor.initializer)
# Get the variable value.
print('global_step: %s' % tf.train.global_step(sess, global_step_tensor))
global_step: 10
```
Args:
sess: A TensorFlow `Session` object.
global_step_tensor: `Tensor` or the `name` of the operation that contains
the global step.
Returns:
The global step value.
"""
if context.executing_eagerly():
return int(global_step_tensor.numpy())
return int(sess.run(global_step_tensor))
@tf_export('train.get_global_step')
def get_global_step(graph=None):
"""Get the global step tensor.
The global step tensor must be an integer variable. We first try to find it
in the collection `GLOBAL_STEP`, or by name `global_step:0`.
Args:
graph: The graph to find the global step in. If missing, use default graph.
Returns:
The global step variable, or `None` if none was found.
Raises:
TypeError: If the global step tensor has a non-integer type, or if it is not
a `Variable`.
"""
graph = graph or ops.get_default_graph()
global_step_tensor = None
global_step_tensors = graph.get_collection(ops.GraphKeys.GLOBAL_STEP)
if len(global_step_tensors) == 1:
global_step_tensor = global_step_tensors[0]
elif not global_step_tensors:
try:
global_step_tensor = graph.get_tensor_by_name('global_step:0')
except KeyError:
return None
else:
logging.error('Multiple tensors in global_step collection.')
return None
assert_global_step(global_step_tensor)
return global_step_tensor
@tf_export('train.create_global_step')
def create_global_step(graph=None):
"""Create global step tensor in graph.
Args:
graph: The graph in which to create the global step tensor. If missing,
use default graph.
Returns:
Global step tensor.
Raises:
ValueError: if global step tensor is already defined.
"""
graph = graph or ops.get_default_graph()
if get_global_step(graph) is not None:
raise ValueError('"global_step" already exists.')
if context.executing_eagerly():
with ops.device('cpu:0'):
return variable_scope.get_variable(
ops.GraphKeys.GLOBAL_STEP,
shape=[],
dtype=dtypes.int64,
initializer=init_ops.zeros_initializer(),
trainable=False,
collections=[ops.GraphKeys.GLOBAL_VARIABLES,
ops.GraphKeys.GLOBAL_STEP])
# Create in proper graph and base name_scope.
with graph.as_default() as g, g.name_scope(None):
return variable_scope.get_variable(
ops.GraphKeys.GLOBAL_STEP,
shape=[],
dtype=dtypes.int64,
initializer=init_ops.zeros_initializer(),
trainable=False,
collections=[ops.GraphKeys.GLOBAL_VARIABLES,
ops.GraphKeys.GLOBAL_STEP])
@tf_export('train.get_or_create_global_step')
def get_or_create_global_step(graph=None):
"""Returns and create (if necessary) the global step tensor.
Args:
graph: The graph in which to create the global step tensor. If missing, use
default graph.
Returns:
The global step tensor.
"""
graph = graph or ops.get_default_graph()
global_step_tensor = get_global_step(graph)
if global_step_tensor is None:
global_step_tensor = create_global_step(graph)
return global_step_tensor
@tf_export('train.assert_global_step')
def assert_global_step(global_step_tensor):
"""Asserts `global_step_tensor` is a scalar int `Variable` or `Tensor`.
Args:
global_step_tensor: `Tensor` to test.
"""
if not (isinstance(global_step_tensor, variables.Variable) or
isinstance(global_step_tensor, ops.Tensor) or
resource_variable_ops.is_resource_variable(global_step_tensor)):
raise TypeError(
'Existing "global_step" must be a Variable or Tensor: %s.' %
global_step_tensor)
if not global_step_tensor.dtype.base_dtype.is_integer:
raise TypeError('Existing "global_step" does not have integer type: %s' %
global_step_tensor.dtype)
if (global_step_tensor.get_shape().ndims != 0 and
global_step_tensor.get_shape().is_fully_defined()):
raise TypeError('Existing "global_step" is not scalar: %s' %
global_step_tensor.get_shape())
def _get_global_step_read(graph=None):
"""Gets global step read tensor in graph.
Args:
graph: The graph in which to create the global step read tensor. If missing,
use default graph.
Returns:
Global step read tensor.
Raises:
RuntimeError: if multiple items found in collection GLOBAL_STEP_READ_KEY.
"""
graph = graph or ops.get_default_graph()
global_step_read_tensors = graph.get_collection(GLOBAL_STEP_READ_KEY)
if len(global_step_read_tensors) > 1:
raise RuntimeError('There are multiple items in collection {}. '
'There should be only one.'.format(GLOBAL_STEP_READ_KEY))
if len(global_step_read_tensors) == 1:
return global_step_read_tensors[0]
return None
def _get_or_create_global_step_read(graph=None):
"""Gets or creates global step read tensor in graph.
Args:
graph: The graph in which to create the global step read tensor. If missing,
use default graph.
Returns:
Global step read tensor if there is global_step_tensor else return None.
"""
graph = graph or ops.get_default_graph()
global_step_read_tensor = _get_global_step_read(graph)
if global_step_read_tensor is not None:
return global_step_read_tensor
global_step_tensor = get_global_step(graph)
if global_step_tensor is None:
return None
# add 'zero' so that it will create a copy of variable as Tensor.
with graph.as_default() as g, g.name_scope(None):
with g.name_scope(global_step_tensor.op.name + '/'):
# using initialized_value to ensure that global_step is initialized before
# this run. This is needed for example Estimator makes all model_fn build
# under global_step_read_tensor dependency.
global_step_value = global_step_tensor.initialized_value() if isinstance(
global_step_tensor, variables.Variable) else global_step_tensor
global_step_read_tensor = global_step_value + 0
ops.add_to_collection(GLOBAL_STEP_READ_KEY, global_step_read_tensor)
return _get_global_step_read(graph)
def _increment_global_step(increment, graph=None):
graph = graph or ops.get_default_graph()
global_step_tensor = get_global_step(graph)
if global_step_tensor is None:
raise ValueError(
'Global step tensor should be created by '
'tf.train.get_or_create_global_step before calling increment.')
global_step_read_tensor = _get_or_create_global_step_read(graph)
with graph.as_default() as g, g.name_scope(None):
with g.name_scope(global_step_tensor.op.name + '/'):
with ops.control_dependencies([global_step_read_tensor]):
return state_ops.assign_add(global_step_tensor, increment)
|
