-
Notifications
You must be signed in to change notification settings - Fork 20
/
Copy pathmnist_fcn.py
184 lines (157 loc) · 7.25 KB
/
mnist_fcn.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
# import MNIST data, Tensorflow, and other helpers
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/")
import tensorflow as tf
import numpy as np
import sys
import os
import pickle
# training parameters
training_epochs = 50
batch_size = 128
# architecture parameters
n_hidden = 128
n_labels = 10
image_pixels = 28 * 28
try:
nonlinearity_name = sys.argv[1] # 'relu', 'elu', 'gelu', 'silu'
except:
print('Defaulted to gelu since no nonlinearity specified through command line')
nonlinearity_name = 'gelu'
try:
learning_rate = float(sys.argv[2]) # 0.001, 0.0001, 0.00001
except:
print('Defaulted to a learning rate of 0.001')
learning_rate = 1e-3
try:
p = float(sys.argv[3]) # 1 or 0.5
except:
print('Defaulted to to a dropout keep probability of 1.0')
p = 1.
x = tf.placeholder(dtype=tf.float32, shape=[None, image_pixels])
y = tf.placeholder(dtype=tf.int64, shape=[None])
is_training = tf.placeholder(tf.bool)
W = {
'1': tf.Variable(tf.nn.l2_normalize(tf.random_normal([image_pixels, n_hidden]), 0)),
'2': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'3': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'4': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'5': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'6': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'7': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'8': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_hidden]), 0)),
'9': tf.Variable(tf.nn.l2_normalize(tf.random_normal([n_hidden, n_labels]), 0))
}
b = {
'1': tf.Variable(tf.zeros([n_hidden])),
'2': tf.Variable(tf.zeros([n_hidden])),
'3': tf.Variable(tf.zeros([n_hidden])),
'4': tf.Variable(tf.zeros([n_hidden])),
'5': tf.Variable(tf.zeros([n_hidden])),
'6': tf.Variable(tf.zeros([n_hidden])),
'7': tf.Variable(tf.zeros([n_hidden])),
'8': tf.Variable(tf.zeros([n_hidden])),
'9': tf.Variable(tf.zeros([n_labels]))
}
def feedforward(x):
if nonlinearity_name == 'relu':
f = tf.nn.relu
elif nonlinearity_name == 'elu':
f = tf.nn.elu
elif nonlinearity_name == 'gelu':
# def gelu(x):
# return tf.mul(x, tf.erfc(-x / tf.sqrt(2.)) / 2.)
# f = gelu
def gelu_fast(_x):
return 0.5 * _x * (1 + tf.tanh(tf.sqrt(2 / np.pi) * (_x + 0.044715 * tf.pow(_x, 3))))
f = gelu_fast
elif nonlinearity_name == 'silu':
def silu(_x):
return _x * tf.sigmoid(_x)
f = silu
# elif nonlinearity_name == 'soi':
# def soi_map(x):
# u = tf.random_uniform(tf.shape(x))
# mask = tf.to_float(tf.less(u, (1 + tf.erf(x / tf.sqrt(2.))) / 2.))
# return tf.cond(is_training, lambda: tf.mul(mask, x),
# lambda: tf.mul(x, tf.erfc(-x / tf.sqrt(2.)) / 2.))
# f = soi_map
else:
raise NameError("Need 'relu', 'elu', 'gelu', or 'silu' for nonlinearity_name")
h1 = f(tf.matmul(x, W['1']) + b['1'])
h1 = tf.cond(is_training, lambda: tf.nn.dropout(h1, p), lambda: h1)
h2 = f(tf.matmul(h1, W['2']) + b['2'])
h2 = tf.cond(is_training, lambda: tf.nn.dropout(h2, p), lambda: h2)
h3 = f(tf.matmul(h2, W['3']) + b['3'])
h3 = tf.cond(is_training, lambda: tf.nn.dropout(h3, p), lambda: h3)
h4 = f(tf.matmul(h3, W['4']) + b['4'])
h4 = tf.cond(is_training, lambda: tf.nn.dropout(h4, p), lambda: h4)
h5 = f(tf.matmul(h4, W['5']) + b['5'])
h5 = tf.cond(is_training, lambda: tf.nn.dropout(h5, p), lambda: h5)
h6 = f(tf.matmul(h5, W['6']) + b['6'])
h6 = tf.cond(is_training, lambda: tf.nn.dropout(h6, p), lambda: h6)
h7 = f(tf.matmul(h6, W['7']) + b['7'])
h7 = tf.cond(is_training, lambda: tf.nn.dropout(h7, p), lambda: h7)
h8 = f(tf.matmul(h7, W['8']) + b['8'])
h8 = tf.cond(is_training, lambda: tf.nn.dropout(h8, p), lambda: h8)
return tf.matmul(h8, W['9']) + b['9']
logits = feedforward(x)
loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits, y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss)
compute_error = tf.reduce_mean(tf.to_float(tf.not_equal(tf.argmax(logits, 1), y)))
# store future results with previous results
if not os.path.exists("./data/"):
os.makedirs("./data/")
if os.path.exists("./data/mnist_fcn_" + nonlinearity_name + ".p"):
history = pickle.load(open("./data/mnist_fcn_" + nonlinearity_name + ".p", "rb"))
key_str = str(len(history)//8 + 1)
history["lr" + key_str] = learning_rate
history["dropout" + key_str] = p
history["train_loss" + key_str] = []
history["val_loss" + key_str] = []
history["test_loss" + key_str] = []
history["train_err" + key_str] = []
history["val_err" + key_str] = []
history["test_err" + key_str] = []
else:
history = {
"lr1": learning_rate, "dropout1": p,
'train_loss1': [], 'val_loss1': [], 'test_loss1': [],
'train_err1': [], 'val_err1': [], 'test_err1': []
}
key_str = '1'
with tf.Session() as sess:
print('Beginning training')
sess.run(tf.initialize_all_variables())
num_batches = mnist.train.num_examples // batch_size
save_every = num_batches//3 # save training information 3 times per epoch
for epoch in range(training_epochs):
for i in range(num_batches):
bx, by = mnist.train.next_batch(batch_size)
if p < 1-1e-5: # we want to know how the full network is being optimized instead of the reduced version
l, err = sess.run([loss, compute_error], feed_dict={x: bx, y: by, is_training: False})
_, l_drop, err_drop = sess.run([optimizer, loss, compute_error], feed_dict={x: bx, y: by,
is_training: True})
if p < 1-1e-5: # we want to know how the full network is being optimized instead of the reduced version
history["train_loss" + key_str].append(l)
history["train_err" + key_str].append(err)
else:
history["train_loss" + key_str].append(l_drop)
history["train_err" + key_str].append(err_drop)
# save
if i % save_every == 0:
l, err = sess.run([loss, compute_error],
feed_dict={x: mnist.validation.images,
y: mnist.validation.labels,
is_training: False})
history["val_loss" + key_str].append(l)
history["val_err" + key_str].append(err)
l, err = sess.run([loss, compute_error],
feed_dict={x: mnist.test.images,
y: mnist.test.labels,
is_training: False})
history["test_loss" + key_str].append(l)
history["test_err" + key_str].append(err)
# print('Epoch', epoch + 1, 'Complete')
# save history
pickle.dump(history, open("./data/mnist_fcn_" + nonlinearity_name + ".p", "wb"))