TF之AE：AE实现TF自带数据集数字真实值对比AE先encoder后decoder预测数字的精确对比—daidingdaiding

1. import tensorflow as tf
2. import numpy as np
3. import matplotlib.pyplot as plt
4. 
5. #Import MNIST data
6. from tensorflow.examples.tutorials.mnist import input_data
7. mnist=input_data.read_data_sets("/niu/mnist_data/",one_hot=False)
8. 
9. 
10. # Parameter
11. learning_rate = 0.01
12. training_epochs = 10
13. batch_size = 256
14. display_step = 1
15. examples_to_show = 10
16. 
17. # Network Parameters
18. n_input = 784
19. 
20. #tf Graph input(only pictures)
21. X=tf.placeholder("float", [None,n_input])
22. 
23. # hidden layer settings
24. n_hidden_1 = 256
25. n_hidden_2 = 128 <br>
26. weights = {
27. 'encoder_h1':tf.Variable(tf.random_normal([n_input,n_hidden_1])),
28. 'encoder_h2': tf.Variable(tf.random_normal([n_hidden_1,n_hidden_2])),
29. 'decoder_h1': tf.Variable(tf.random_normal([n_hidden_2,n_hidden_1])),
30. 'decoder_h2': tf.Variable(tf.random_normal([n_hidden_1, n_input])),
31.     }
32. biases = {
33. 'encoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),
34. 'encoder_b2': tf.Variable(tf.random_normal([n_hidden_2])),
35. 'decoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),
36. 'decoder_b2': tf.Variable(tf.random_normal([n_input])),
37.     }
38. 
39. #定义encoder
40. def encoder(x):
41. # Encoder Hidden layer with sigmoid activation #1
42.     layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['encoder_h1']),
43.                                    biases['encoder_b1']))
44. # Decoder Hidden layer with sigmoid activation #2
45.     layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['encoder_h2']),
46.                                    biases['encoder_b2']))
47. return layer_2
48. 
49. #定义decoder
50. def decoder(x):
51. # Encoder Hidden layer with sigmoid activation #1
52.     layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['decoder_h1']),
53.                                    biases['decoder_b1']))
54. # Decoder Hidden layer with sigmoid activation #2
55.     layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['decoder_h2']),
56.                                    biases['decoder_b2']))
57. return layer_2
58. 
59. # Construct model
60. encoder_op = encoder(X)             # 128 Features
61. decoder_op = decoder(encoder_op)    # 784 Features
62. 
63. # Prediction
64. y_pred = decoder_op   
65. # Targets (Labels) are the input data.
66. y_true = X            
67. 
68. # Define loss and optimizer, minimize the squared error
69. 
70. cost = tf.reduce_mean(tf.pow(y_true - y_pred, 2))
71. optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)
72. 
73. # Launch the graph
74. with tf.Session() as sess:<br>
75.     sess.run(tf.initialize_all_variables())
76.     total_batch = int(mnist.train.num_examples/batch_size)
77. # Training cycle
78. for epoch in range(training_epochs):
79. # Loop over all batches
80. for i in range(total_batch):
81.             batch_xs, batch_ys = mnist.train.next_batch(batch_size)  # max(x) = 1, min(x) = 0
82. # Run optimization op (backprop) and cost op (to get loss value)
83.             _, c = sess.run([optimizer, cost], feed_dict={X: batch_xs})
84. # Display logs per epoch step
85. if epoch % display_step == 0:
86. print("Epoch:", '%04d' % (epoch+1),
87. "cost=", "{:.9f}".format(c))
88. 
89. print("Optimization Finished!")
90. # # Applying encode and decode over test set
91.     encode_decode = sess.run(
92.         y_pred, feed_dict={X: mnist.test.images[:examples_to_show]})
93. # Compare original images with their reconstructions
94.     f, a = plt.subplots(2, 10, figsize=(10, 2))
95.     plt.title('Matplotlib,AE--Jason Niu')
96. for i in range(examples_to_show):
97.         a[0][i].imshow(np.reshape(mnist.test.images[i], (28, 28)))
98.         a[1][i].imshow(np.reshape(encode_decode[i], (28, 28)))
99.     plt.show()