forked from google-deepmind/learning-to-learn
-
Notifications
You must be signed in to change notification settings - Fork 0
/
networks_test.py
189 lines (160 loc) · 6.54 KB
/
networks_test.py
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
# Copyright 2016 Google Inc.
#
# 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
#
# https://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 L2L networks."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from nose_parameterized import parameterized
import numpy as np
import sonnet as snt
import tensorflow as tf
import networks
class CoordinateWiseDeepLSTMTest(tf.test.TestCase):
"""Tests CoordinateWiseDeepLSTM network."""
def testShape(self):
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.CoordinateWiseDeepLSTM(layers=(1, 1))
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
self.assertEqual(update.get_shape().as_list(), shape)
def testTrainable(self):
"""Tests the network contains trainable variables."""
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.CoordinateWiseDeepLSTM(layers=(1,))
state = net.initial_state_for_inputs(gradients)
net(gradients, state)
# Weights and biases for two layers.
variables = snt.get_variables_in_module(net)
self.assertEqual(len(variables), 4)
@parameterized.expand([
["zeros"],
[{"w": "zeros", "b": "zeros", "bad": "bad"}],
[{"w": tf.zeros_initializer(), "b": np.array([0])}],
[{"linear": {"w": tf.zeros_initializer(), "b": "zeros"}}]
])
def testResults(self, initializer):
"""Tests zero updates when last layer is initialized to zero."""
shape = [10]
gradients = tf.random_normal(shape)
net = networks.CoordinateWiseDeepLSTM(layers=(1, 1),
initializer=initializer)
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
update_np = sess.run(update)
self.assertAllEqual(update_np, np.zeros(shape))
class KernelDeepLSTMTest(tf.test.TestCase):
"""Tests KernelDeepLSTMTest network."""
def testShape(self):
kernel_shape = [5, 5]
shape = kernel_shape + [2, 2] # The input has to be 4-dimensional.
gradients = tf.random_normal(shape)
net = networks.KernelDeepLSTM(layers=(1, 1), kernel_shape=kernel_shape)
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
self.assertEqual(update.get_shape().as_list(), shape)
def testTrainable(self):
"""Tests the network contains trainable variables."""
kernel_shape = [5, 5]
shape = kernel_shape + [2, 2] # The input has to be 4-dimensional.
gradients = tf.random_normal(shape)
net = networks.KernelDeepLSTM(layers=(1,), kernel_shape=kernel_shape)
state = net.initial_state_for_inputs(gradients)
net(gradients, state)
# Weights and biases for two layers.
variables = snt.get_variables_in_module(net)
self.assertEqual(len(variables), 4)
@parameterized.expand([
["zeros"],
[{"w": "zeros", "b": "zeros", "bad": "bad"}],
[{"w": tf.zeros_initializer(), "b": np.array([0])}],
[{"linear": {"w": tf.zeros_initializer(), "b": "zeros"}}]
])
def testResults(self, initializer):
"""Tests zero updates when last layer is initialized to zero."""
kernel_shape = [5, 5]
shape = kernel_shape + [2, 2] # The input has to be 4-dimensional.
gradients = tf.random_normal(shape)
net = networks.KernelDeepLSTM(layers=(1, 1),
kernel_shape=kernel_shape,
initializer=initializer)
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
update_np = sess.run(update)
self.assertAllEqual(update_np, np.zeros(shape))
class SgdTest(tf.test.TestCase):
"""Tests Sgd network."""
def testShape(self):
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.Sgd()
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
self.assertEqual(update.get_shape().as_list(), shape)
def testNonTrainable(self):
"""Tests the network doesn't contain trainable variables."""
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.Sgd()
state = net.initial_state_for_inputs(gradients)
net(gradients, state)
variables = snt.get_variables_in_module(net)
self.assertEqual(len(variables), 0)
def testResults(self):
"""Tests network produces zero updates with learning rate equal to zero."""
shape = [10]
learning_rate = 0.01
gradients = tf.random_normal(shape)
net = networks.Sgd(learning_rate=learning_rate)
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
with self.test_session() as sess:
gradients_np, update_np = sess.run([gradients, update])
self.assertAllEqual(update_np, -learning_rate * gradients_np)
class AdamTest(tf.test.TestCase):
"""Tests Adam network."""
def testShape(self):
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.Adam()
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
self.assertEqual(update.get_shape().as_list(), shape)
def testNonTrainable(self):
"""Tests the network doesn't contain trainable variables."""
shape = [10, 5]
gradients = tf.random_normal(shape)
net = networks.Adam()
state = net.initial_state_for_inputs(gradients)
net(gradients, state)
variables = snt.get_variables_in_module(net)
self.assertEqual(len(variables), 0)
def testZeroLearningRate(self):
"""Tests network produces zero updates with learning rate equal to zero."""
shape = [10]
gradients = tf.random_normal(shape)
net = networks.Adam(learning_rate=0)
state = net.initial_state_for_inputs(gradients)
update, _ = net(gradients, state)
with self.test_session() as sess:
update_np = sess.run(update)
self.assertAllEqual(update_np, np.zeros(shape))
if __name__ == "__main__":
tf.test.main()