-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathLSTM-DenseNet.py
175 lines (146 loc) · 6.63 KB
/
LSTM-DenseNet.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
import tensorflow as tf
import time
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
sess = tf.InteractiveSession()
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
def rnn_layer(x, timesteps, num_hidden, weights, bias):
x = tf.unstack(x, timesteps, 1)
lstm_cell_a = tf.contrib.rnn.BasicLSTMCell(num_hidden, forget_bias=1.0)
lstm_cell_b = tf.contrib.rnn.BasicLSTMCell(num_hidden, forget_bias=1.0)
outputs, _, _ = tf.contrib.rnn.static_bidirectional_rnn(
lstm_cell_a, lstm_cell_b, x, dtype=tf.float32)
return tf.nn.relu(tf.matmul(outputs[-1], weights) + bias)
# Placeholders
x = tf.placeholder(tf.float32, shape=[None, 784])
y_ = tf.placeholder(tf.float32, shape=[None, 10])
# Main ConvLayer with max-pooling
with tf.variable_scope('ConvLayer'):
W_conv = weight_variable([5, 5, 1, 64])
b_conv = bias_variable([64])
x_image = tf.reshape(x, [-1, 28, 28, 1])
h_conv = tf.nn.relu(conv2d(x_image, W_conv) + b_conv)
h_pool = max_pool_2x2(h_conv)
with tf.variable_scope('ConvLayer1'):
# ConvLayer 1
W_conv1 = weight_variable([3, 3, 64, 64])
b_conv1 = bias_variable([64])
h_conv1 = tf.nn.relu(conv2d(h_pool, W_conv1) + b_conv1)
with tf.variable_scope('ConvLayer2'):
# ConvLayer 2
W_conv2 = weight_variable([3, 3, 64, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2) + b_conv2)
with tf.variable_scope('ConvLayer3'):
# ConvLayer 3
W_conv3_1 = weight_variable([3, 3, 64, 64])
b_conv3_1 = bias_variable([64])
h_conv3_1 = tf.nn.relu(conv2d(h_conv1, W_conv3_1) + b_conv3_1)
W_conv3 = weight_variable([3, 3, 64, 64])
b_conv3 = bias_variable([64])
h_conv3 = tf.nn.relu(conv2d(h_conv2 + h_conv3_1, W_conv3) + b_conv3)
with tf.variable_scope('ConvLayer4'):
# ConvLayer 4
W_conv4_1 = weight_variable([3, 3, 64, 64])
b_conv4_1 = bias_variable([64])
h_conv4_1 = tf.nn.relu(conv2d(h_conv1, W_conv4_1) + b_conv4_1)
W_conv4_2 = weight_variable([3, 3, 64, 64])
b_conv4_2 = bias_variable([64])
h_conv4_2 = tf.nn.relu(conv2d(h_conv2, W_conv4_2) + b_conv4_2)
W_conv4 = weight_variable([3, 3, 64, 64])
b_conv4 = bias_variable([64])
h_conv4 = tf.nn.relu(conv2d(h_conv3 + h_conv4_1 + h_conv4_2, W_conv4) + b_conv4)
with tf.variable_scope('ConvLayer5'):
# ConvLayer 5
W_conv5_1 = weight_variable([3, 3, 64, 64])
b_conv5_1 = bias_variable([64])
h_conv5_1 = tf.nn.relu(conv2d(h_conv1, W_conv5_1) + b_conv5_1)
W_conv5_2 = weight_variable([3, 3, 64, 64])
b_conv5_2 = bias_variable([64])
h_conv5_2 = tf.nn.relu(conv2d(h_conv2, W_conv5_2) + b_conv5_2)
W_conv5_3 = weight_variable([3, 3, 64, 64])
b_conv5_3 = bias_variable([64])
h_conv5_3 = tf.nn.relu(conv2d(h_conv3, W_conv5_3) + b_conv5_3)
W_conv5 = weight_variable([3, 3, 64, 64])
b_conv5 = bias_variable([64])
h_conv5 = tf.nn.relu(conv2d(h_conv4 + h_conv5_1 + h_conv5_2 + h_conv5_3, W_conv5) + b_conv5)
with tf.variable_scope('ConvLayer6'):
# ConvLayer 6
W_conv6_1 = weight_variable([3, 3, 64, 64])
b_conv6_1 = bias_variable([64])
h_conv6_1 = tf.nn.relu(conv2d(h_conv1, W_conv6_1) + b_conv6_1)
W_conv6_2 = weight_variable([3, 3, 64, 64])
b_conv6_2 = bias_variable([64])
h_conv6_2 = tf.nn.relu(conv2d(h_conv2, W_conv6_2) + b_conv6_2)
W_conv6_3 = weight_variable([3, 3, 64, 64])
b_conv6_3 = bias_variable([64])
h_conv6_3 = tf.nn.relu(conv2d(h_conv3, W_conv6_3) + b_conv6_3)
W_conv6_4 = weight_variable([3, 3, 64, 64])
b_conv6_4 = bias_variable([64])
h_conv6_4 = tf.nn.relu(conv2d(h_conv4, W_conv6_4) + b_conv6_4)
W_conv6 = weight_variable([3, 3, 64, 64])
b_conv6 = bias_variable([64])
h_conv6 = tf.nn.relu(conv2d(h_conv5 + h_conv6_1 + h_conv6_2 + h_conv6_3 + h_conv6_4, W_conv6) + b_conv6)
with tf.variable_scope('LSTM'):
# LSTM Layer
h_conv6_reshape = tf.reshape(tf.transpose(h_conv6, [0, 3, 1, 2]), [-1, 64, 14 * 14])
W_lstm = weight_variable([1024 * 2, 2048])
b_lstm = bias_variable([2048])
h_lstm = rnn_layer(h_conv6_reshape, 64, 1024, W_lstm, b_lstm)
# Dropout
keep_prob = tf.placeholder(tf.float32)
h_dense1_drop = tf.nn.dropout(h_lstm, keep_prob)
# Dense Layer 2 with Softmax Output
W_dense2 = weight_variable([2048, 10])
b_dense2 = bias_variable([10])
y_conv = tf.matmul(h_dense1_drop, W_dense2) + b_dense2
# Training Parameters
training_rate = tf.placeholder(tf.float32)
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
train_step = tf.train.AdamOptimizer(training_rate).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
saver = tf.train.Saver(tf.global_variables())
with tf.Session() as sess:
# Training
sess.run(tf.global_variables_initializer())
data_location = './Mixed-LSTM_ConvNet-DATA/MNIST'
#saver.restore(sess, data_location)
last_time = time.time()
rate = 0.0001
for i in range(100000):
batch = mnist.train.next_batch(50)
sess.run(train_step, feed_dict={
x: batch[0], y_: batch[1], keep_prob: 0.5, training_rate: rate})
if i % 10 == 0:
loss, acc = sess.run([cross_entropy, accuracy], feed_dict={
x: batch[0], y_: batch[1], keep_prob: 1.0, training_rate: rate})
print('Step: %d, Accuracy: %.2f, Loss: %.5f, Speed: %.1f sec/10 steps' %
(i, acc, loss, time.time() - last_time))
last_time = time.time()
if i % 1000 == 0 and i > 0:
current_accuracy = accuracy.eval(
feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0, training_rate: rate})
print('- Current Test Accuracy %.4f' % current_accuracy)
saver.save(sess, data_location)
print('- Model Saved in Step %d' % i)
if current_accuracy > 0.98:
rate = 0.00003
if current_accuracy > 0.99:
rate = 0.00001
if current_accuracy > 0.993:
rate = 0.000003
if current_accuracy > 0.995:
print('- Accuracy Reached 99.5% in Step %d' % i)
break
last_time = time.time()