#!/usr/bin/python
# -*- coding: utf-8 -*-
"""A Variational Autoencoders for MNIST.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from keras.layers import Input, Dense, Lambda, Conv2D, Conv2DTranspose, \
Flatten, Reshape
from keras.models import Model
from keras import backend as K
from keras.datasets import mnist
from keras import metrics
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from scipy.stats import norm
EPOCH = 5
INPUT_DIM = 784
BATCH_SIZE = 64
HIDDEN_VAR_DIM = 7 * 7 * 32
LATENT_VAR_DIM = 2
# input image dimensions
(img_rows, img_cols, img_chns) = (28, 28, 1)
if K.image_data_format() == 'channels_first':
original_img_size = (img_chns, img_rows, img_cols)
output_shape = (BATCH_SIZE, 32, 7, 7)
else:
original_img_size = (img_rows, img_cols, img_chns)
output_shape = (BATCH_SIZE, 7, 7, 32)
def sampling(args):
(z_mean, z_var) = args
epsilon = K.random_normal(shape=(K.shape(z_mean)[0],
LATENT_VAR_DIM), mean=0., stddev=1.)
return z_mean + z_var * epsilon
def encode(x):
input_reshape = Reshape(original_img_size)(x)
conv1 = Conv2D(16, 5, strides=(2, 2), padding='same',
activation='relu')(input_reshape)
conv2 = Conv2D(32, 5, strides=(2, 2), padding='same',
activation='relu')(conv1)
hidden = Flatten()(conv2)
z_mean = Dense(LATENT_VAR_DIM, activation='relu')(hidden)
z_var = Dense(LATENT_VAR_DIM, activation='relu')(hidden)
return (z_mean, z_var)
def decode(z):
hidden = Dense(HIDDEN_VAR_DIM, activation='relu')(z)
hidden_reshape = Reshape(output_shape[1:])(hidden)
deconv1 = Conv2DTranspose(16, 5, strides=(2, 2), padding='same',
activation='relu')(hidden_reshape)
deconv2 = Conv2DTranspose(1, 5, strides=(2, 2), padding='same',
activation='sigmoid')(deconv1)
return Flatten()(deconv2)
def main(_):
x = Input(shape=(INPUT_DIM, ))
(z_mean, z_var) = encode(x)
z = Lambda(sampling)([z_mean, z_var])
x_decoded = decode(z)
model = Model(inputs=x, outputs=x_decoded)
def vae_loss(y_true, y_pred):
generation_loss = img_rows * img_cols \
* metrics.binary_crossentropy(x, x_decoded)
kl_loss = 0.5 * tf.reduce_sum(K.square(z_mean)
+ K.square(z_var) - K.log(K.square(z_var + 1e-8)) - 1,
axis=1)
return tf.reduce_mean(generation_loss + kl_loss)
model.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:])))
print(model.summary())
model.fit(
x_train,
y_train,
shuffle=True,
epochs=EPOCH,
batch_size=BATCH_SIZE,
validation_data=(x_test, y_test),
)
generator = K.function([model.layers[8].input],
[model.layers[12].output])
# 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))
# linearly spaced coordinates on the unit square were transformed through the inverse CDF (ppf) of the Gaussian
# to produce values of the latent variables z, since the prior of the latent space is Gaussian
grid_x = norm.ppf(np.linspace(0.05, 0.95, n))
grid_y = norm.ppf(np.linspace(0.05, 0.95, n))
for (i, yi) in enumerate(grid_x):
for (j, xi) in enumerate(grid_y):
z_sample = np.array([[xi, yi]])
z_sample = np.tile(z_sample,
BATCH_SIZE).reshape(BATCH_SIZE, 2)
x_decoded = generator([z_sample])[0]
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, cmap='Greys_r')
plt.show()
if __name__ == '__main__':
tf.app.run(main=main)
