1
2 import numpy as np
3 import tensorflow as tf
4 import matplotlib
5 import matplotlib.pyplot as plt
6 import matplotlib.cm as cm
7 from tensorflow.examples.tutorials.mnist import input_data
8
9
10 # 训练的准确度目标
11 accuracyGoal = 0.98
12
13 # 是否已经达到指定的准确度
14 bFlagGoal = False;
15
16 # 显示数字的图像,nBytes为784个点的灰度值，浮点数
17 def showMnistImg(nBytes):
18 imgBytes = nBytes.reshape((28, 28))
19 #print(imgBytes)
20 plt.figure(figsize=(2.8,2.8))
21 #plt.grid() #开启网格
22 plt.imshow(imgBytes, cmap=cm.gray)
23 plt.show()
24
25
26 #加载mnist数据
27 mnist = input_data.read_data_sets('MNIST_data', one_hot=True)
28
29 ### 单个手写数字的784个点的灰度值，浮点数，范围[0,1)
30 ##print('type(mnist.train.images[0]): ', type(mnist.train.images[0])) # <class 'numpy.ndarray'>
31 ##print('mnist.train.images.shape: ', mnist.train.images.shape)
32 ##print(mnist.train.images[0])
33 ##
34 ##
35 ### 单个手写数字的标签
36 ### 一个one-hot向量除了某一位的数字是1以外其余各维度数字都是0
37 ### 数字n将表示成一个只有在第n维度（从0开始）数字为1的10维向量。
38 ##print('type(mnist.train.labels[0]): ', type(mnist.train.labels[0]))# <class 'numpy.ndarray'>
39 ##print('type(mnist.train.labels.shape): ', type(mnist.train.labels.shape))
40 ##print(mnist.train.labels[0])
41
42
43
44 # 下面开始CNN相关
45
46 def conv2d(x, W):
47 return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
48
49 def max_pool_2x2(x):
50 return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
51 strides=[1, 2, 2, 1], padding='SAME')
52
53
54 def weight_variable(shape):
55 initial = tf.truncated_normal(shape, stddev=0.1)
56 return tf.Variable(initial)
57
58 def bias_variable(shape):
59 initial = tf.constant(0.1, shape=shape)
60 return tf.Variable(initial)
61
62
63 x = tf.placeholder(tf.float32, shape=[None, 784])
64 y_ = tf.placeholder(tf.float32, shape=[None, 10])
65
66
67 W_conv1 = weight_variable([5, 5, 1, 32])
68 b_conv1 = bias_variable([32])
69
70 x_image = tf.reshape(x, [-1, 28, 28, 1])
71
72 h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
73 h_pool1 = max_pool_2x2(h_conv1)
74
75
76 W_conv2 = weight_variable([5, 5, 32, 64])
77 b_conv2 = bias_variable([64])
78
79 h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
80 h_pool2 = max_pool_2x2(h_conv2)
81
82
83
84 W_fc1 = weight_variable([7 * 7 * 64, 1024])
85 b_fc1 = bias_variable([1024])
86
87 h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])
88 h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
89
90
91 keep_prob = tf.placeholder(tf.float32)
92 h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
93
94
95 W_fc2 = weight_variable([1024, 10])
96 b_fc2 = bias_variable([10])
97
98 y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
99
100
101 cross_entropy = tf.reduce_mean(
102 tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_, logits=y_conv))
103 train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
104 correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
105 accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
106
107
108
109
110 print('\n开始训练...')
111 with tf.Session() as sess:
112 sess.run(tf.global_variables_initializer())
113 for i in range(3000):
114 batch = mnist.train.next_batch(50)
115
116 if i % 100 == 0:
117 train_accuracy = accuracy.eval(feed_dict={ x: batch[0], y_: batch[1], keep_prob: 1.0})
118 print('次数 %d, 准确度 %g' % (i, train_accuracy))
119
120 if(train_accuracy>accuracyGoal):
121 #创建saver对象，它添加了一些op用来save和restore模型参数
122 saver = tf.train.Saver()
123 #使用saver提供的简便方法去调用 save op
124 saver.save(sess, "saved_model/cnn_handwrite_number.ckpt")
125
126 print('已保存模型')
127 bFlagGoal = True
128 break
129
130 train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
131
132 if(bFlagGoal):
133 print('训练结束，已达到训练目标')
134 else:
135 print('训练结束，未完成训练目标')
136
137
138
139