SAE（稀疏编码器）源码备用

利用Python实现SAE（稀疏编码器），输入为TensorFlow封装的mnist手写识别数据集

# coding=utf-8
'''
Created on 2016年12月3日
@author: chunsoft
'''
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
# 导入 MNIST 数据
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

# 参数
learning_rate = 0.01  # 学习速率
training_epochs = 20  # 训练批次
batch_size = 256  # 随机选择训练数据大小
display_step = 1  # 展示步骤
examples_to_show = 10  # 显示示例图片数量

# 网络参数
# 我这里采用了三层编码，实际针对mnist数据，隐层两层，分别为256，128效果最好
n_hidden_1 = 512  # 第一隐层神经元数量
n_hidden_2 = 256  # 第二
n_hidden_3 = 128  # 第三
n_input = 784  # 输入

# tf Graph输入
X = tf.placeholder("float", [None, n_input])

# 权重初始化
weights = {
    'encoder_h1': tf.Variable(tf.random_normal([n_input, n_hidden_1])),
    'encoder_h2': tf.Variable(tf.random_normal([n_hidden_1, n_hidden_2])),
    'encoder_h3': tf.Variable(tf.random_normal([n_hidden_2, n_hidden_3])),
    'decoder_h1': tf.Variable(tf.random_normal([n_hidden_3, n_hidden_2])),
    'decoder_h2': tf.Variable(tf.random_normal([n_hidden_2, n_hidden_1])),
    'decoder_h3': tf.Variable(tf.random_normal([n_hidden_1, n_input])),
}

# 偏置值初始化
biases = {
    'encoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),
    'encoder_b2': tf.Variable(tf.random_normal([n_hidden_2])),
    'encoder_b3': tf.Variable(tf.random_normal([n_hidden_3])),
    'decoder_b1': tf.Variable(tf.random_normal([n_hidden_2])),
    'decoder_b2': tf.Variable(tf.random_normal([n_hidden_1])),
    'decoder_b3': tf.Variable(tf.random_normal([n_input])),
}


# 开始编码
def encoder(x):
    # sigmoid激活函数，layer = x*weights['encoder_h1']+biases['encoder_b1']
    layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['encoder_h1']),
                                   biases['encoder_b1']))
    layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['encoder_h2']),
                                   biases['encoder_b2']))
    layer_3 = tf.nn.sigmoid(tf.add(tf.matmul(layer_2, weights['encoder_h3']),
                                   biases['encoder_b3']))
    return layer_3


# 开始解码
def decoder(x):
    # sigmoid激活函数,layer = x*weights['decoder_h1']+biases['decoder_b1']
    layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['decoder_h1']),
                                   biases['decoder_b1']))
    layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['decoder_h2']),
                                   biases['decoder_b2']))
    layer_3 = tf.nn.sigmoid(tf.add(tf.matmul(layer_2, weights['decoder_h3']),
                                   biases['decoder_b3']))
    return layer_3


# 构造模型
encoder_op = encoder(X)
encoder_result = encoder_op
decoder_op = decoder(encoder_op)

# 预测
y_pred = decoder_op
# 实际输入数据当作标签
y_true = X

# 定义代价函数和优化器，最小化平方误差,这里可以根据实际修改误差模型
cost = tf.reduce_mean(tf.pow(y_true - y_pred, 2))
optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(cost)

# 初始化变量
init = tf.initialize_all_variables();

# 运行Graph
with tf.Session() as sess:
    sess.run(init)
    # 总的batch
    total_batch = int(mnist.train.num_examples / batch_size)
    # 开始训练
    for epoch in range(training_epochs):
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            _, c = sess.run([optimizer, cost], feed_dict={X: batch_xs})
        # 展示每次训练结果
        if epoch % display_step == 0:
            print("Epoch:", '%04d' % (epoch + 1),
                  "cost=", "{:.9f}".format(c))
    print("Optimization Finished!")
    # Applying encode and decode over test set
    # 显示编码结果和解码后结果
    encodes = sess.run(
        encoder_result, feed_dict={X: mnist.test.images[:examples_to_show]})
    encode_decode = sess.run(
        y_pred, feed_dict={X: mnist.test.images[:examples_to_show]})
    # 对比原始图片重建图片
    f, a = plt.subplots(2, 10, figsize=(10, 2))
    for i in range(examples_to_show):
        a[0][i].imshow(np.reshape(mnist.test.images[i], (28, 28)))
        a[1][i].imshow(np.reshape(encode_decode[i], (28, 28)))
    f.show()
    plt.draw()
    plt.waitforbuttonpress()