TensorFlow教程06：MNIST的CNN实现——源码和运行结果

假定您已经安装好了TensorFlow，这里放了第二个MNIST实验的代码和参考结果，你可以直接运行验证。

源码

[python] view plain copy

print?

1. #!/usr/bin/python  
2. import tensorflow as tf  
3. import sys  
4. from tensorflow.examples.tutorials.mnist import input_data  
5.   
6. def weight_variable(shape):  
7.   initial = tf.truncated_normal(shape, stddev=0.1)  
8.   return tf.Variable(initial)  
9.   
10. def bias_variable(shape):  
11.   initial = tf.constant(0.1, shape=shape)  
12.   return tf.Variable(initial)  
13.   
14. def conv2d(x, W):  
15.   return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')  
16.   
17. def max_pool_2x2(x):  
18.   return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')  
19.   
20. mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)  
21.   
22. sess = tf.InteractiveSession()  
23.   
24. x = tf.placeholder("float", shape=[None, 784])  
25. y_ = tf.placeholder("float", shape=[None, 10])  
26.   
27. W = tf.Variable(tf.zeros([784,10]))  
28. b = tf.Variable(tf.zeros([10]))  
29.   
30. W_conv1 = weight_variable([5, 5, 1, 32])  
31. b_conv1 = bias_variable([32])  
32.   
33. x_image = tf.reshape(x, [-1, 28, 28, 1])  
34.   
35. h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)  
36. h_pool1 = max_pool_2x2(h_conv1)  
37.   
38. W_conv2 = weight_variable([5, 5, 32, 64])  
39. b_conv2 = bias_variable([64])  
40.   
41. h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)  
42. h_pool2 = max_pool_2x2(h_conv2)  
43.   
44. # Now image size is reduced to 7*7  
45. W_fc1 = weight_variable([7 * 7 * 64, 1024])  
46. b_fc1 = bias_variable([1024])  
47.   
48. h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])  
49. h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)  
50.   
51. keep_prob = tf.placeholder("float")  
52. h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)  
53.   
54. W_fc2 = weight_variable([1024, 10])  
55. b_fc2 = bias_variable([10])  
56. y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)  
57.   
58.   
59. cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))  
60. train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)  
61. correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))  
62. accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))  
63. sess.run(tf.initialize_all_variables())  
64.   
65. for i in range(20000):  
66.   batch = mnist.train.next_batch(50)  
67.   if i%100 == 0:  
68.     train_accuracy = accuracy.eval(feed_dict={  
69.         x:batch[0], y_: batch[1], keep_prob: 1.0})  
70.     print "step %d, training accuracy %.3f"%(i, train_accuracy)  
71.   train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})  
72.   
73. print "Training finished"  
74.   
75. print "test accuracy %.3f" % accuracy.eval(feed_dict={  
76.     x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})  

#!/usr/bin/pythonimport tensorflow as tfimport sysfrom tensorflow.examples.tutorials.mnist import input_datadef 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')mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)sess = tf.InteractiveSession()x = tf.placeholder("float", shape=[None, 784])y_ = tf.placeholder("float", shape=[None, 10])W = tf.Variable(tf.zeros([784,10]))b = tf.Variable(tf.zeros([10]))W_conv1 = weight_variable([5, 5, 1, 32])b_conv1 = bias_variable([32])x_image = tf.reshape(x, [-1, 28, 28, 1])h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)h_pool1 = max_pool_2x2(h_conv1)W_conv2 = weight_variable([5, 5, 32, 64])b_conv2 = bias_variable([64])h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)h_pool2 = max_pool_2x2(h_conv2)# Now image size is reduced to 7*7W_fc1 = weight_variable([7 * 7 * 64, 1024])b_fc1 = bias_variable([1024])h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)keep_prob = tf.placeholder("float")h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)W_fc2 = weight_variable([1024, 10])b_fc2 = bias_variable([10])y_conv=tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)cross_entropy = -tf.reduce_sum(y_*tf.log(y_conv))train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)correct_prediction = tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))sess.run(tf.initialize_all_variables())for i in range(20000):  batch = mnist.train.next_batch(50)  if i%100 == 0:    train_accuracy = accuracy.eval(feed_dict={        x:batch[0], y_: batch[1], keep_prob: 1.0})    print "step %d, training accuracy %.3f"%(i, train_accuracy)  train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})print "Training finished"print "test accuracy %.3f" % accuracy.eval(feed_dict={    x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})

运行结果

[plain] view plain copy

print?

1. Extracting MNIST_data/train-images-idx3-ubyte.gz  
2. Extracting MNIST_data/train-labels-idx1-ubyte.gz  
3. Extracting MNIST_data/t10k-images-idx3-ubyte.gz  
4. Extracting MNIST_data/t10k-labels-idx1-ubyte.gz  
5. step 0, training accuracy 0.140  
6. step 100, training accuracy 0.840  
7. step 200, training accuracy 0.900  
8. step 300, training accuracy 0.840  
9. step 400, training accuracy 0.980  
10. step 500, training accuracy 0.940  
11. ...  
12. step 19900, training accuracy 1.000  
13. Training finished  
14. test accuracy 0.993  

Extracting MNIST_data/train-images-idx3-ubyte.gzExtracting MNIST_data/train-labels-idx1-ubyte.gzExtracting MNIST_data/t10k-images-idx3-ubyte.gzExtracting MNIST_data/t10k-labels-idx1-ubyte.gzstep 0, training accuracy 0.140step 100, training accuracy 0.840step 200, training accuracy 0.900step 300, training accuracy 0.840step 400, training accuracy 0.980step 500, training accuracy 0.940...step 19900, training accuracy 1.000Training finishedtest accuracy 0.993