TF入门之自编码器

参考资料&＃xff1a;
https://github.com/aymericdamien/TensorFlow-Examples/blob/master/examples/3_NeuralNetworks/autoencoder.py

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as pltfrom tensorflow.examples.tutorials.mnist import input_data
mnist &＃61; input_data.read_data_sets("/tmp/data/", one_hot&＃61;True)"""
自编码器将输入经过多层网络计算得到一个中间状态&＃xff0c;
然后再将这个中间状态经过多层网络计算得到还原状态&＃xff0c;
计算还原的状态和原始状态之间的误差&＃xff0c;从而更新网络层的权重
"""learning_rate &＃61; 0.01
num_steps &＃61; 20000
batch_size &＃61; 256display_step &＃61; 1000
examples_to_show &＃61; 10num_hidden_1 &＃61; 256
num_hidden_2 &＃61; 128
num_input &＃61; 784X &＃61; tf.placeholder(&＃39;float&＃39;, [None, num_input])"""
input -> encoder_layer1 -> encoder_layer2 -> decoder_layer1 -> decoder_layer2 -> output
"""weights &＃61; {&＃39;encoder_h1&＃39;: tf.Variable(tf.random_normal([num_input, num_hidden_1])),&＃39;encoder_h2&＃39;: tf.Variable(tf.random_normal([num_hidden_1, num_hidden_2])),&＃39;decoder_h1&＃39;: tf.Variable(tf.random_normal([num_hidden_2, num_hidden_1])),&＃39;decoder_h2&＃39;: tf.Variable(tf.random_normal([num_hidden_1, num_input])),
}biases &＃61; {&＃39;encoder_b1&＃39;: tf.Variable(tf.random_normal([num_hidden_1])),&＃39;encoder_b2&＃39;: tf.Variable(tf.random_normal([num_hidden_2])),&＃39;decoder_b1&＃39;: tf.Variable(tf.random_normal([num_hidden_1])),&＃39;decoder_b2&＃39;: tf.Variable(tf.random_normal([num_input])),
}def encoder(x):layer1 &＃61; tf.nn.sigmoid(tf.add(tf.matmul(x,weights["encoder_h1"]), biases["encoder_b1"]))layer2 &＃61; tf.nn.sigmoid(tf.add(tf.matmul(layer1, weights["encoder_h2"]), biases["encoder_b2"]))return layer2
def decoder(x):layer1 &＃61; tf.nn.sigmoid(tf.add(tf.matmul(x, weights["decoder_h1"]), biases["decoder_b1"]))layer2 &＃61; tf.nn.sigmoid(tf.add(tf.matmul(layer1, weights["decoder_h2"]), biases["decoder_b2"]))return layer2encoder_op &＃61; encoder(X)
decoder_op &＃61; decoder(encoder_op)y_pred &＃61; decoder_op
y_true &＃61; X#计算误差
loss &＃61; tf.reduce_mean(tf.pow(y_true-y_pred,2))
#优化器
optimizer &＃61; tf.train.RMSPropOptimizer(learning_rate).minimize(loss)init &＃61; tf.global_variables_initializer()with tf.Session() as sess:sess.run(init)for i in range(1, num_steps&＃43;1):batch_x, _ &＃61; mnist.train.next_batch(batch_size)_, l &＃61; sess.run([optimizer, loss], feed_dict&＃61;{X:batch_x})if i % display_step &＃61;&＃61; 0 or i &＃61;&＃61; 1:print("loss", l)n &＃61; 1for i in range(10):batch_x, _ &＃61; mnist.test.next_batch(1)g &＃61; sess.run(decoder_op, feed_dict&＃61;{X:batch_x})g &＃61; np.reshape(g, newshape&＃61;[28, 28])print("Original Images")plt.figure(figsize&＃61;(n, n))plt.imshow(np.reshape(batch_x, newshape&＃61;[28, 28]), origin&＃61;"upper", cmap&＃61;"gray")plt.show()print("Reconstructed Images")plt.figure(figsize&＃61;(n, n))plt.imshow(g, origin&＃61;"upper", cmap&＃61;"gray")plt.show()