TF之AE：AE實作TF自帶資料集數字真實值對比AE先encoder後decoder預測數字的精确對比

輸出結果

代碼設計

import tensorflow as tf

import numpy as np

import matplotlib.pyplot as plt

#Import MNIST data

from tensorflow.examples.tutorials.mnist import input_data

mnist=input_data.read_data_sets("/niu/mnist_data/",one_hot=False)

# Parameter

learning_rate = 0.01

training_epochs = 10

batch_size = 256

display_step = 1

examples_to_show = 10

# Network Parameters

n_input = 784

#tf Graph input(only pictures)

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

# hidden layer settings

n_hidden_1 = 256

n_hidden_2 = 128 <br>

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])),

   'decoder_h1': tf.Variable(tf.random_normal([n_hidden_2,n_hidden_1])),

   'decoder_h2': tf.Variable(tf.random_normal([n_hidden_1, n_input])),

   }

biases = {

   'encoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),

   'encoder_b2': tf.Variable(tf.random_normal([n_hidden_2])),

   'decoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),

   'decoder_b2': tf.Variable(tf.random_normal([n_input])),

#定義encoder

def encoder(x):

   # Encoder Hidden layer with sigmoid activation #1

   layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['encoder_h1']),

                                  biases['encoder_b1']))

   # Decoder Hidden layer with sigmoid activation #2

   layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['encoder_h2']),

                                  biases['encoder_b2']))

   return layer_2

#定義decoder

def decoder(x):

   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']))

# Construct model

encoder_op = encoder(X)             # 128 Features

decoder_op = decoder(encoder_op)    # 784 Features

# Prediction

y_pred = decoder_op  

# Targets (Labels) are the input data.

y_true = X            

# Define loss and optimizer, minimize the squared error

cost = tf.reduce_mean(tf.pow(y_true - y_pred, 2))

optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)

# Launch the graph

with tf.Session() as sess:<br>

   sess.run(tf.initialize_all_variables())

   total_batch = int(mnist.train.num_examples/batch_size)

   # Training cycle

   for epoch in range(training_epochs):

       # Loop over all batches

       for i in range(total_batch):

           batch_xs, batch_ys = mnist.train.next_batch(batch_size)  # max(x) = 1, min(x) = 0

           # Run optimization op (backprop) and cost op (to get loss value)

           _, c = sess.run([optimizer, cost], feed_dict={X: batch_xs})

       # Display logs per epoch step

       if epoch % display_step == 0:

           print("Epoch:", '%04d' % (epoch+1),

                 "cost=", "{:.9f}".format(c))

   print("Optimization Finished!")

   # # Applying encode and decode over test set

   encode_decode = sess.run(

       y_pred, feed_dict={X: mnist.test.images[:examples_to_show]})

   # Compare original images with their reconstructions

   f, a = plt.subplots(2, 10, figsize=(10, 2))

   plt.title('Matplotlib,AE--Jason Niu')

   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)))

   plt.show()