10.variational_autoencoder

程序说明

时间：2016年11月17日

说明：该程序构造一个变分自动编码器。

Reference: "Auto-Encoding Variational Bayes"

数据集：MNIST

1.加载keras模块

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
from keras.layers import Input, Dense, Lambda
from keras.models import Model
from keras import backend as K
from keras import objectives
from keras.datasets import mnist

Using TensorFlow backend.

2.变量初始化

batch_size = 100
original_dim = 784
latent_dim = 2
intermediate_dim = 256
nb_epoch = 50
epsilon_std = 1.0

x = Input(batch_shape=(batch_size, original_dim))
h = Dense(intermediate_dim, activation='relu')(x)
z_mean = Dense(latent_dim)(h)
z_log_var = Dense(latent_dim)(h)

def sampling(args):
    z_mean, z_log_var = args
    epsilon = K.random_normal(shape=(batch_size, latent_dim), mean=0.,
                              std=epsilon_std)
    return z_mean + K.exp(z_log_var / 2) * epsilon

# note that "output_shape" isn't necessary with the TensorFlow backend
z = Lambda(sampling, output_shape=(latent_dim,))([z_mean, z_log_var])

# we instantiate these layers separately so as to reuse them later
decoder_h = Dense(intermediate_dim, activation='relu')
decoder_mean = Dense(original_dim, activation='sigmoid')
h_decoded = decoder_h(z)
x_decoded_mean = decoder_mean(h_decoded)

def vae_loss(x, x_decoded_mean):
    xent_loss = original_dim * objectives.binary_crossentropy(x, x_decoded_mean)
    kl_loss = - 0.5 * K.sum(1 + z_log_var - K.square(z_mean) - K.exp(z_log_var), axis=-1)
    return xent_loss + kl_loss

vae = Model(x, x_decoded_mean)
vae.compile(optimizer='rmsprop', loss=vae_loss)

# train the VAE on MNIST digits
(x_train, y_train), (x_test, y_test) = mnist.load_data()

x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))

vae.fit(x_train, x_train,
        shuffle=True,
        nb_epoch=nb_epoch,
        batch_size=batch_size,
        validation_data=(x_test, x_test))

# build a model to project inputs on the latent space
encoder = Model(x, z_mean)

Train on 60000 samples, validate on 10000 samples
Epoch 1/50
60000/60000 [==============================] - 5s - loss: 191.0905 - val_loss: 173.5436
Epoch 2/50
60000/60000 [==============================] - 5s - loss: 171.7320 - val_loss: 169.0859
Epoch 3/50
60000/60000 [==============================] - 5s - loss: 168.1224 - val_loss: 166.7636
Epoch 4/50
60000/60000 [==============================] - 5s - loss: 165.5420 - val_loss: 164.4730
Epoch 5/50
60000/60000 [==============================] - 5s - loss: 163.8741 - val_loss: 163.4501
Epoch 6/50
60000/60000 [==============================] - 5s - loss: 162.7379 - val_loss: 162.9273
Epoch 7/50
60000/60000 [==============================] - 5s - loss: 161.8719 - val_loss: 161.7938
Epoch 8/50
60000/60000 [==============================] - 5s - loss: 161.0796 - val_loss: 161.3434
Epoch 9/50
60000/60000 [==============================] - 5s - loss: 160.4377 - val_loss: 160.4363
Epoch 10/50
60000/60000 [==============================] - 5s - loss: 159.8330 - val_loss: 159.9203
Epoch 11/50
60000/60000 [==============================] - 5s - loss: 159.2621 - val_loss: 159.5222
Epoch 12/50
60000/60000 [==============================] - 5s - loss: 158.7406 - val_loss: 158.9181
Epoch 13/50
60000/60000 [==============================] - 5s - loss: 158.2449 - val_loss: 158.6521
Epoch 14/50
60000/60000 [==============================] - 5s - loss: 157.7862 - val_loss: 158.1587
Epoch 15/50
60000/60000 [==============================] - 5s - loss: 157.3053 - val_loss: 157.8226
Epoch 16/50
60000/60000 [==============================] - 5s - loss: 156.9417 - val_loss: 158.0445
Epoch 17/50
60000/60000 [==============================] - 5s - loss: 156.5658 - val_loss: 157.3171
Epoch 18/50
60000/60000 [==============================] - 5s - loss: 156.2249 - val_loss: 157.0271
Epoch 19/50
60000/60000 [==============================] - 5s - loss: 155.8873 - val_loss: 157.0536
Epoch 20/50
60000/60000 [==============================] - 5s - loss: 155.5819 - val_loss: 156.3859
Epoch 21/50
60000/60000 [==============================] - 5s - loss: 155.2945 - val_loss: 156.1629
Epoch 22/50
60000/60000 [==============================] - 5s - loss: 155.0148 - val_loss: 155.9180
Epoch 23/50
60000/60000 [==============================] - 5s - loss: 154.7706 - val_loss: 155.9890
Epoch 24/50
60000/60000 [==============================] - 5s - loss: 154.5395 - val_loss: 155.7588
Epoch 25/50
60000/60000 [==============================] - 5s - loss: 154.3486 - val_loss: 155.7783
Epoch 26/50
60000/60000 [==============================] - 5s - loss: 154.1120 - val_loss: 155.4132
Epoch 27/50
60000/60000 [==============================] - 5s - loss: 153.9194 - val_loss: 155.1289
Epoch 28/50
60000/60000 [==============================] - 5s - loss: 153.6825 - val_loss: 155.4338
Epoch 29/50
60000/60000 [==============================] - 5s - loss: 153.5506 - val_loss: 155.0366
Epoch 30/50
60000/60000 [==============================] - 5s - loss: 153.3196 - val_loss: 154.7996
Epoch 31/50
60000/60000 [==============================] - 5s - loss: 153.1628 - val_loss: 154.6084
Epoch 32/50
60000/60000 [==============================] - 4s - loss: 152.9938 - val_loss: 155.1787
Epoch 33/50
60000/60000 [==============================] - 5s - loss: 152.8034 - val_loss: 154.5734
Epoch 34/50
60000/60000 [==============================] - 5s - loss: 152.6532 - val_loss: 154.3676
Epoch 35/50
60000/60000 [==============================] - 4s - loss: 152.4781 - val_loss: 154.4870
Epoch 36/50
60000/60000 [==============================] - 4s - loss: 152.3254 - val_loss: 154.2274
Epoch 37/50
60000/60000 [==============================] - 4s - loss: 152.1693 - val_loss: 154.5915
Epoch 38/50
60000/60000 [==============================] - 4s - loss: 151.9978 - val_loss: 154.0650
Epoch 39/50
60000/60000 [==============================] - 4s - loss: 151.8608 - val_loss: 153.7670
Epoch 40/50
60000/60000 [==============================] - 4s - loss: 151.7393 - val_loss: 153.9039
Epoch 41/50
60000/60000 [==============================] - 4s - loss: 151.5966 - val_loss: 154.7774
Epoch 42/50
60000/60000 [==============================] - 5s - loss: 151.4757 - val_loss: 153.7938
Epoch 43/50
60000/60000 [==============================] - 5s - loss: 151.3705 - val_loss: 153.8829
Epoch 44/50
60000/60000 [==============================] - 5s - loss: 151.2328 - val_loss: 153.7716
Epoch 45/50
60000/60000 [==============================] - 4s - loss: 151.1133 - val_loss: 153.3561
Epoch 46/50
60000/60000 [==============================] - 4s - loss: 151.0194 - val_loss: 153.4168
Epoch 47/50
60000/60000 [==============================] - 4s - loss: 150.8907 - val_loss: 153.4800
Epoch 48/50
60000/60000 [==============================] - 4s - loss: 150.7891 - val_loss: 153.5535
Epoch 49/50
60000/60000 [==============================] - 4s - loss: 150.6946 - val_loss: 153.8616
Epoch 50/50
60000/60000 [==============================] - 4s - loss: 150.5713 - val_loss: 153.4389

# display a 2D plot of the digit classes in the latent space
x_test_encoded = encoder.predict(x_test, batch_size=batch_size)
plt.figure(figsize=(6, 6))
plt.scatter(x_test_encoded[:, 0], x_test_encoded[:, 1], c=y_test)
plt.colorbar()
plt.show()

# build a digit generator that can sample from the learned distribution
decoder_input = Input(shape=(latent_dim,))
_h_decoded = decoder_h(decoder_input)
_x_decoded_mean = decoder_mean(_h_decoded)
generator = Model(decoder_input, _x_decoded_mean)

# display a 2D manifold of the digits
n = 15  # figure with 15x15 digits
digit_size = 28
figure = np.zeros((digit_size * n, digit_size * n))
# we will sample n points within [-15, 15] standard deviations
grid_x = np.linspace(-15, 15, n)
grid_y = np.linspace(-15, 15, n)

for i, yi in enumerate(grid_x):
    for j, xi in enumerate(grid_y):
        z_sample = np.array([[xi, yi]])
        x_decoded = generator.predict(z_sample)
        digit = x_decoded[0].reshape(digit_size, digit_size)
        figure[i * digit_size: (i + 1) * digit_size,
               j * digit_size: (j + 1) * digit_size] = digit

plt.figure(figsize=(10, 10))
plt.imshow(figure)
plt.show()