06_autoencoder.py

from keras.models import Model
from keras.layers import Dense, Input
from keras.datasets import mnist
import numpy as np

# Hyper parameters
batch_size = 128
nb_epoch = 100

# Parameters for MNIST dataset
img_rows, img_cols = 28, 28

# Parameters for denoising autoencoder
nb_visible = img_rows * img_cols
nb_hidden = 500
corruption_level = 0.3

# Load MNIST Dataset
(x_train, _), (x_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(x_train.shape)
print(x_test.shape)

# Add random noise
x_train_noisy = x_train + corruption_level * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape)
x_test_noisy = x_test + corruption_level * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape)
x_train_noisy = np.clip(x_train_noisy, 0., 1.)
x_test_noisy = np.clip(x_test_noisy, 0., 1.)
print(x_train_noisy.shape)
print(x_test_noisy.shape)

# Build autoencoder model
input_img = Input(shape=(nb_visible,))
encoded = Dense(nb_hidden, activation='relu')(input_img)
decoded = Dense(nb_visible, activation='sigmoid')(encoded)

autoencoder = Model(input=input_img, output=decoded)
autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')
autoencoder.summary()

# Train
autoencoder.fit(x_train_noisy, x_train,
                nb_epoch=nb_epoch, batch_size=batch_size, shuffle=True, verbose=1,
                validation_data=(x_test_noisy, x_test))

# Evaluate
evaluation = autoencoder.evaluate(x_test_noisy, x_test, batch_size=batch_size, verbose=1)
print('\nSummary: Loss over the test dataset: %.2f' % evaluation)