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
|
# 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.
# ==============================================================================
"""RMSprop optimizer for Tensorflow.
rmsprop algorithm [tieleman2012rmsprop]
A detailed description of rmsprop.
- maintain a moving (discounted) average of the square of gradients
- divide gradient by the root of this average
mean_square = decay * mean_square{t-1} + (1-decay) * gradient ** 2
mom = momentum * mom{t-1} + learning_rate * g_t / sqrt(mean_square + epsilon)
delta = - mom
This implementation of RMSProp uses plain momentum, not Nesterov momentum.
The centered version additionally maintains a moving (discounted) average of the
gradients, and uses that average to estimate the variance:
mean_grad = decay * mean_square{t-1} + (1-decay) * gradient
mean_square = decay * mean_square{t-1} + (1-decay) * gradient ** 2
mom = momentum * mom{t-1} + learning_rate * g_t /
sqrt(mean_square - mean_grad**2 + epsilon)
delta = - mom
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.contrib.optimizer_v2 import optimizer_v2
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import init_ops
from tensorflow.python.training import training_ops
class RMSPropOptimizer(optimizer_v2.OptimizerV2):
"""Optimizer that implements the RMSProp algorithm.
See the
[paper](http://www.cs.toronto.edu/~tijmen/csc321/slides/lecture_slides_lec6.pdf).
"""
def __init__(self,
learning_rate,
decay=0.9,
momentum=0.0,
epsilon=1e-10,
use_locking=False,
centered=False,
name="RMSProp"):
"""Construct a new RMSProp optimizer.
Note that in the dense implementation of this algorithm, variables and their
corresponding accumulators (momentum, gradient moving average, square
gradient moving average) will be updated even if the gradient is zero
(i.e. accumulators will decay, momentum will be applied). The sparse
implementation (used when the gradient is an `IndexedSlices` object,
typically because of `tf.gather` or an embedding lookup in the forward pass)
will not update variable slices or their accumulators unless those slices
were used in the forward pass (nor is there an "eventual" correction to
account for these omitted updates). This leads to more efficient updates for
large embedding lookup tables (where most of the slices are not accessed in
a particular graph execution), but differs from the published algorithm.
Some of the args below are hyperparameters, where a hyperparameter is
defined as a scalar Tensor, a regular Python value or a callable (which
will be evaluated when `apply_gradients` is called) returning a scalar
Tensor or a Python value.
Args:
learning_rate: A float hyperparameter. The learning rate.
decay: A float hyperparameter. Discounting factor for the history/coming
gradient.
momentum: A float hyperparameter.
epsilon: A float hyperparameter. Small value to avoid zero denominator.
use_locking: If True use locks for update operation.
centered: If True, gradients are normalized by the estimated variance of
the gradient; if False, by the uncentered second moment. Setting this to
True may help with training, but is slightly more expensive in terms of
computation and memory. Defaults to False.
name: Optional name prefix for the operations created when applying
gradients. Defaults to "RMSProp".
"""
super(RMSPropOptimizer, self).__init__(use_locking, name)
self._set_hyper("learning_rate", learning_rate)
self._set_hyper("decay", decay)
self._set_hyper("momentum", momentum)
self._set_hyper("epsilon", epsilon)
self._centered = centered
def _create_vars(self, var_list, state):
for v in var_list:
if v.get_shape().is_fully_defined():
init_rms = init_ops.ones_initializer(dtype=v.dtype.base_dtype)
else:
init_rms = array_ops.ones_like(v)
state.create_slot_with_initializer(v, init_rms, v.get_shape(),
v.dtype.base_dtype, "rms")
if self._centered:
state.zeros_slot(v, "mg")
state.zeros_slot(v, "momentum")
def _apply_dense(self, grad, var, state):
rms = state.get_slot(var, "rms")
mom = state.get_slot(var, "momentum")
if self._centered:
mg = state.get_slot(var, "mg")
return training_ops.apply_centered_rms_prop(
var,
mg,
rms,
mom,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
use_locking=self._use_locking).op
else:
return training_ops.apply_rms_prop(
var,
rms,
mom,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
use_locking=self._use_locking).op
def _resource_apply_dense(self, grad, var, state):
rms = state.get_slot(var, "rms")
mom = state.get_slot(var, "momentum")
if self._centered:
mg = state.get_slot(var, "mg")
return training_ops.resource_apply_centered_rms_prop(
var.handle,
mg.handle,
rms.handle,
mom.handle,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
use_locking=self._use_locking)
else:
return training_ops.resource_apply_rms_prop(
var.handle,
rms.handle,
mom.handle,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
use_locking=self._use_locking)
def _apply_sparse(self, grad, var, state):
rms = state.get_slot(var, "rms")
mom = state.get_slot(var, "momentum")
if self._centered:
mg = state.get_slot(var, "mg")
return training_ops.sparse_apply_centered_rms_prop(
var,
mg,
rms,
mom,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad.values,
grad.indices,
use_locking=self._use_locking)
else:
return training_ops.sparse_apply_rms_prop(
var,
rms,
mom,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad.values,
grad.indices,
use_locking=self._use_locking)
def _resource_apply_sparse(self, grad, var, indices, state):
rms = state.get_slot(var, "rms")
mom = state.get_slot(var, "momentum")
if self._centered:
mg = self.get_slot(var, "mg")
return training_ops.resource_sparse_apply_centered_rms_prop(
var.handle,
mg.handle,
rms.handle,
mom.handle,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
indices,
use_locking=self._use_locking)
else:
return training_ops.resource_sparse_apply_rms_prop(
var.handle,
rms.handle,
mom.handle,
state.get_hyper("learning_rate", var.dtype.base_dtype),
state.get_hyper("decay", var.dtype.base_dtype),
state.get_hyper("momentum", var.dtype.base_dtype),
state.get_hyper("epsilon", var.dtype.base_dtype),
grad,
indices,
use_locking=self._use_locking)
|
