-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathlenet5_ori.py
152 lines (126 loc) · 5.5 KB
/
lenet5_ori.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
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import sys
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
import numpy as np
FLAGS = None
save_file='./ckptlenet/last.ckpt'
#输入：32x32
#第1层：卷积 6 5x5 -> 28x28x6
#第2层：池化 2x2 -> 14x14x6
#第3层：卷积 16 5x5 -> 10x10x16
#第4层：池化 2x2 -> 5x5x16
#第5层：全连接 120 -> 120
#第6层：全连接 84 ->84
#第7层：全连接 10 ->10
#6位整数，10位小数
#定义一堆函数
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='VALID')
def conv2d_same(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
"""max_pool_2x2 downsamples a feature map by 2X."""
# 2x2的池化
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1] ,
strides=[1, 2, 2, 1], padding='VALID')
def weight_variable(shape,name):
"""weight_variable generates a weight variable of a given shape."""
# 产生一个初始化的正态分布的权重张量，均值是0，标准差是0.1，而且权重不会大于2倍的标准差
initial = tf.truncated_normal(shape,dtype=tf.float32, stddev=0.1)
# 返回图变量，该图变量具有初始值initial
return tf.Variable(initial,name=name)
def bias_variable(shape,name):
"""bias_variable generates a bias variable of a given shape."""
# 所有的偏移均为0.1
initial = tf.constant(0.1, shape=shape,dtype=tf.float32)
return tf.Variable(initial,name=name)
print("import data")
# Import data
mnist = input_data.read_data_sets("../MNIST_data", one_hot=True)
# Create the model
# 定义输入层
# Define loss and optimizer
# 定义输出结果
with tf.variable_scope('input'):
x = tf.placeholder(tf.float32, [None, 784],name='x_input')
y_ = tf.placeholder(tf.float32, [None, 10],name='y_output')
keep_prob = tf.placeholder(tf.float32)
with tf.variable_scope('converse'):
x_image = tf.reshape(x, [-1, 28, 28, 1])
#定义第一层卷积
#第1层：卷积 6 5x5 -> 28x28x6
with tf.variable_scope('Layer1'):
W_conv1 = weight_variable([5, 5, 1, 6],'L1_weight')
b_conv1 = bias_variable([6],'L1_bias')
conv1_result = conv2d_same(x_image, W_conv1)
result_l1 = tf.nn.relu(conv1_result + b_conv1)
#第2层：池化 2x2 -> 14x14x6
with tf.variable_scope('Layer2'):
result_l2 = max_pool_2x2(result_l1)
#第3层：卷积 16 5x5 -> 10x10x16
with tf.variable_scope('Layer3'):
W_conv2 = weight_variable([5, 5, 6, 16],'L3_weight')
b_conv2 = bias_variable([16],'L3_bias')
conv2_result = conv2d(result_l2, W_conv2)
result_l3 = tf.nn.relu(conv2_result + b_conv2)
#第4层：池化 2x2 -> 5x5x16
with tf.variable_scope('Layer4'):
result_l4 = max_pool_2x2(result_l3)
#第5层：全连接 120 -> 120
result_l4_flat = tf.reshape(result_l4, [-1, 5*5*16])
with tf.variable_scope('Layer5'):
W_fc1 = weight_variable([5 * 5 * 16, 120],'F1_weight')
b_fc1 = bias_variable([120],'F1_bias')
fc1_result = tf.matmul(result_l4_flat, W_fc1)
fc1_result_relu = tf.nn.relu(fc1_result + b_fc1)
result_l5 = tf.nn.dropout(fc1_result_relu,keep_prob)
#第6层：全连接 84 ->84
with tf.variable_scope('Layer6'):
W_fc2 = weight_variable([120, 84],'F2_weight')
b_fc2 = bias_variable([84],'F2_bias')
fc2_result = tf.matmul(result_l5, W_fc2)
fc2_result_relu = tf.nn.relu(fc2_result + b_fc2)
result_l6 = tf.nn.dropout(fc2_result_relu,keep_prob)
#第7层：全连接 10 ->10
with tf.variable_scope('Layer7'):
W_fc3 = weight_variable([84, 10],'F3_weight')
b_fc3 = bias_variable([10],'F3_bias')
fc3_result = tf.matmul(result_l6, W_fc3)
result_l7 = tf.nn.relu(fc3_result + b_fc3)
y_conv = result_l7
# 定义每次训练的损失函数和步长
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=y_, logits=y_conv))
#cross_entropy_int16 = tf.cast(cross_entropy*1024,tf.int16)
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
# 正确率判断，同样是搭建环境
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()
fout = open("predict.txt",'w')
print("start train")
with tf.Session() as sess:
# 初始化所有的占位符，之前的函数叫initialize_all_variables，已被下式替代
sess.run(tf.global_variables_initializer())
#batch = mnist.train.next_batch(1)
#print(y_conv.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0}))
for i in range(30000):
batch = mnist.train.next_batch(50)
if i % 100 == 0:
# 每次进行正确性测试，仅使用前50张图片
train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1],keep_prob:1})
print('step %d, training accuracy %g' % (i, train_accuracy))
fout.write("step %d, training accuracy %g\n" % (i, train_accuracy))
# keep_prob 留存率，即dropout时需要训练的节点
train_step.run(feed_dict={x: batch[0], y_: batch[1],keep_prob:0.5})
# 最后的正确性测试，使用全部的图片
print('test accuracy %g' % accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels,keep_prob:1}))
fout.write('test accuracy %g\n' % accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels,keep_prob:1}))
# 保存最后的权重
saver.save(sess,save_file)
print('parameter has saved.')
fout.close()