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
|
# 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.
# ==============================================================================
"""Tests for Proximal Adagrad optimizer."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensorflow.compiler.tests import xla_test
from tensorflow.python.framework import constant_op
from tensorflow.python.ops import resource_variable_ops
from tensorflow.python.ops import variables
from tensorflow.python.platform import test
from tensorflow.python.training import adagrad
from tensorflow.python.training import proximal_adagrad
class ProximalAdagradOptimizerTest(xla_test.XLATestCase):
def testResourceProximalAdagradwithoutRegularization(self):
with self.cached_session(), self.test_scope():
var0 = resource_variable_ops.ResourceVariable([0.0, 0.0])
var1 = resource_variable_ops.ResourceVariable([0.0, 0.0])
grads0 = constant_op.constant([0.1, 0.2])
grads1 = constant_op.constant([0.01, 0.02])
opt = proximal_adagrad.ProximalAdagradOptimizer(
3.0,
initial_accumulator_value=0.1,
l1_regularization_strength=0.0,
l2_regularization_strength=0.0)
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
variables.global_variables_initializer().run()
self.assertAllClose([0.0, 0.0], var0.eval())
self.assertAllClose([0.0, 0.0], var1.eval())
# Run 3 steps Proximal Adagrad.
for _ in range(3):
update.run()
self.assertAllClose(np.array([-2.60260963, -4.29698515]), var0.eval())
self.assertAllClose(np.array([-0.28432083, -0.56694895]), var1.eval())
opt_vars = opt.variables()
self.assertStartsWith(opt_vars[0].name, var0._shared_name)
self.assertStartsWith(opt_vars[1].name, var1._shared_name)
self.assertEqual(2, len(opt_vars))
def testProximalAdagradwithoutRegularization2(self):
with self.cached_session(), self.test_scope():
var0 = resource_variable_ops.ResourceVariable([1.0, 2.0])
var1 = resource_variable_ops.ResourceVariable([4.0, 3.0])
grads0 = constant_op.constant([0.1, 0.2])
grads1 = constant_op.constant([0.01, 0.02])
opt = proximal_adagrad.ProximalAdagradOptimizer(
3.0,
initial_accumulator_value=0.1,
l1_regularization_strength=0.0,
l2_regularization_strength=0.0)
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
variables.global_variables_initializer().run()
self.assertAllClose([1.0, 2.0], var0.eval())
self.assertAllClose([4.0, 3.0], var1.eval())
# Run 3 steps Proximal Adagrad.
for _ in range(3):
update.run()
self.assertAllClose(np.array([-1.60261, -2.296985]), var0.eval())
self.assertAllClose(np.array([3.715679, 2.433051]), var1.eval())
def testProximalAdagradWithL1(self):
with self.cached_session(), self.test_scope():
var0 = resource_variable_ops.ResourceVariable([1.0, 2.0])
var1 = resource_variable_ops.ResourceVariable([4.0, 3.0])
grads0 = constant_op.constant([0.1, 0.2])
grads1 = constant_op.constant([0.01, 0.02])
opt = proximal_adagrad.ProximalAdagradOptimizer(
3.0,
initial_accumulator_value=0.1,
l1_regularization_strength=0.001,
l2_regularization_strength=0.0)
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
variables.global_variables_initializer().run()
self.assertAllClose([1.0, 2.0], var0.eval())
self.assertAllClose([4.0, 3.0], var1.eval())
# Run 10 steps Proximal Adagrad
for _ in range(10):
update.run()
self.assertAllClose(np.array([-6.663634, -9.190331]), var0.eval())
self.assertAllClose(np.array([2.959304, 1.029232]), var1.eval())
def testProximalAdagradWithL1_L2(self):
with self.cached_session(), self.test_scope():
var0 = resource_variable_ops.ResourceVariable([1.0, 2.0])
var1 = resource_variable_ops.ResourceVariable([4.0, 3.0])
grads0 = constant_op.constant([0.1, 0.2])
grads1 = constant_op.constant([0.01, 0.02])
opt = proximal_adagrad.ProximalAdagradOptimizer(
3.0,
initial_accumulator_value=0.1,
l1_regularization_strength=0.001,
l2_regularization_strength=2.0)
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
variables.global_variables_initializer().run()
self.assertAllClose([1.0, 2.0], var0.eval())
self.assertAllClose([4.0, 3.0], var1.eval())
# Run 10 steps Proximal Adagrad.
for _ in range(10):
update.run()
self.assertAllClose(np.array([-0.0495, -0.0995]), var0.eval())
self.assertAllClose(np.array([-0.0045, -0.0095]), var1.eval())
def applyOptimizer(self, opt, steps=5):
var0 = resource_variable_ops.ResourceVariable([1.0, 2.0])
var1 = resource_variable_ops.ResourceVariable([3.0, 4.0])
grads0 = constant_op.constant([0.1, 0.2])
grads1 = constant_op.constant([0.01, 0.02])
update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
variables.global_variables_initializer().run()
self.assertAllClose([1.0, 2.0], var0.eval())
self.assertAllClose([3.0, 4.0], var1.eval())
# Run ProximalAdagrad for a few steps
for _ in range(steps):
update.run()
return var0.eval(), var1.eval()
def testEquivAdagradwithoutRegularization(self):
with self.cached_session(), self.test_scope():
val0, val1 = self.applyOptimizer(
proximal_adagrad.ProximalAdagradOptimizer(
3.0,
initial_accumulator_value=0.1,
l1_regularization_strength=0.0,
l2_regularization_strength=0.0))
with self.cached_session(), self.test_scope():
val2, val3 = self.applyOptimizer(
adagrad.AdagradOptimizer(
3.0, initial_accumulator_value=0.1))
self.assertAllClose(val0, val2)
self.assertAllClose(val1, val3)
if __name__ == "__main__":
test.main()
|
