VAE_in_tensorflow

vae.ipynb

vae.py

import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt

# Import MNIST data
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/home/grt1st/data/", one_hot=True)

# Parameters
learning_rate = 0.001
training_epochs = 101
batch_size = 100
display_step = 5

# Network Parameters
n_hidden_1 = 500
n_hidden_2 = 500
n_z = 20
n_input = 784 # MNIST data input (img shape: 28*28)

# func
transfer_func = tf.nn.relu#tf.nn.softplus

# tf Graph input (only pictures)
X = tf.placeholder(tf.float32, [None, n_input])

def xavier_init(fan_in, fan_out, constant=1):
    """ Xavier initialization of network weights"""
    # https://stackoverflow.com/questions/33640581/how-to-do-xavier-initialization-on-tensorflow
    low = -constant*np.sqrt(6.0/(fan_in + fan_out))
    high = constant*np.sqrt(6.0/(fan_in + fan_out))
    return tf.random_uniform((fan_in, fan_out),
                             minval=low, maxval=high,
                             dtype=tf.float32)

weights = {
    "recognition_h1": tf.Variable(xavier_init(n_input, n_hidden_1)),
    "recognition_h2": tf.Variable(xavier_init(n_hidden_1, n_hidden_2)),
    "recognition_out_mean": tf.Variable(xavier_init(n_hidden_2, n_z)),
    "recognition_out_log": tf.Variable(xavier_init(n_hidden_2, n_z)),
    "generator_h1": tf.Variable(xavier_init(n_z, n_hidden_1)),
    "generator_h2": tf.Variable(xavier_init(n_hidden_1, n_hidden_2)),
    "generator_out_mean": tf.Variable(xavier_init(n_hidden_2, n_input)),
    "generator_out_log": tf.Variable(xavier_init(n_hidden_2, n_input)),
}

biases = {
    "recognition_b1": tf.Variable(tf.zeros([n_hidden_1], dtype=tf.float32)),
    "recognition_b2": tf.Variable(tf.zeros([n_hidden_2], dtype=tf.float32)),
    "recognition_out_mean": tf.Variable(tf.zeros([n_z], dtype=tf.float32)),
    "recognition_out_log": tf.Variable(tf.zeros([n_z], dtype=tf.float32)),
    "generator_b1": tf.Variable(tf.zeros([n_hidden_1], dtype=tf.float32)),
    "generator_b2": tf.Variable(tf.zeros([n_hidden_2], dtype=tf.float32)),
    "generator_out_mean": tf.Variable(tf.zeros([n_input], dtype=tf.float32)),
    "generator_out_log": tf.Variable(tf.zeros([n_input], dtype=tf.float32)),
}

def recognition_network(x):
    layer_1 = transfer_func(tf.add(tf.matmul(x, weights["recognition_h1"]), biases["recognition_b1"]))
    layer_2 = transfer_func(tf.add(tf.matmul(layer_1, weights["recognition_h2"]), biases["recognition_b2"]))

    z_mean = tf.add(tf.matmul(layer_2, weights["recognition_out_mean"]), biases["recognition_out_mean"])
    z_log = tf.add(tf.matmul(layer_2, weights["recognition_out_log"]), biases["recognition_out_log"])

    return z_mean, z_log

def generator_network(x):
    layer_1 = transfer_func(tf.add(tf.matmul(x, weights["generator_h1"]), biases["generator_b1"]))
    layer_2 = transfer_func(tf.add(tf.matmul(layer_1, weights["generator_h2"]), biases["generator_b2"]))

    x_reconstr_mean = tf.nn.sigmoid(tf.add(tf.matmul(layer_2, weights["generator_out_mean"]), biases["generator_out_mean"]))

    return x_reconstr_mean

z_mean, z_log = recognition_network(X)

eps = tf.random_normal((batch_size, n_z), 0, 1, dtype=tf.float32)

# z = mu + sigma*epsilon
z = tf.add(z_mean, tf.multiply(tf.sqrt(tf.exp(z_log)), eps))

x_reconstr_mean = generator_network(z)

# 重建损失：负对数概率，伯努利分布下的输入
reconstr_loss = -tf.reduce_sum(X * tf.log(1e-10 + x_reconstr_mean) + (1 - X) * tf.log(1e-10 + 1 - x_reconstr_mean), 1)

# 潜在损失
latent_loss = -0.5 * tf.reduce_sum(1 + z_log - tf.square(z_mean) - tf.exp(z_log), 1)

cost = tf.reduce_mean(reconstr_loss + latent_loss)

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

'''
# Initializing the tensor flow variables
        init = tf.global_variables_initializer()

        # Launch the session
        self.sess = tf.InteractiveSession()
        self.sess.run(init)
'''

def generate(session, z_mu=None):
    """ Generate data by sampling from latent space.

    If z_mu is not None, data for this point in latent space is
    generated. Otherwise, z_mu is drawn from prior in latent 
    space.        
    """
    if z_mu is None:
        z_mu = np.random.normal(size=n_z)
    # Note: This maps to mean of distribution, we could alternatively
    # sample from Gaussian distribution
    return session.run(x_reconstr_mean, feed_dict={z: z_mu})



# Initializing the variables
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    total_batch = int(mnist.train.num_examples / batch_size)
    # Training cycle
    for epoch in range(training_epochs):
        avg_cost = 0
        # Loop over all batches
        for i in range(total_batch):
            batch_xs, batch_ys = mnist.train.next_batch(batch_size)
            # Run optimization op (backprop) and cost op (to get loss value)
            _, c = sess.run([optimizer, cost], feed_dict={X: batch_xs})
            avg_cost += c / mnist.train.num_examples * batch_size
        # Display logs per epoch step
        if epoch % display_step == 0:
            print("Epoch:", '%04d' % (epoch + 1),
                  "cost=", "{:.9f}".format(c))

    print("Optimization Finished!")

    x_sample = mnist.test.next_batch(100)[0]
    x_reconstruct = sess.run(x_reconstr_mean, feed_dict={X: x_sample})

    plt.figure(figsize=(8, 12))
    for i in range(5):
        plt.subplot(5, 2, 2 * i + 1)
        plt.imshow(x_sample[i].reshape(28, 28), vmin=0, vmax=1, cmap="gray")
        plt.title("Test input")
        plt.colorbar()
        plt.subplot(5, 2, 2 * i + 2)
        plt.imshow(x_reconstruct[i].reshape(28, 28), vmin=0, vmax=1, cmap="gray")
        plt.title("Reconstruction")
        plt.colorbar()
    #plt.tight_layout()
    plt.show()
    plt.draw()