ae_mnist_relu_adadelta.py

import numpy as np
from keraspp.skeras import plot_history
import matplotlib.pyplot as plt
from keras.datasets import mnist
from keras import layers, models


class AE(models.Model):
    def __init__(self, x_nodes=784, z_dim=36):
        x_shape = (x_nodes,)
        self.x = layers.Input(shape=x_shape)
        self.z = layers.Dense(z_dim, activation='relu')(self.x)
        self.y = layers.Dense(x_nodes, activation='sigmoid')(self.z)
        super().__init__(self.x, self.y)
        self.z_dim = z_dim
        print("Compiling model...")
        self.compile(optimizer='adadelta', loss='binary_crossentropy', metrics=['accuracy'])

    def Encoder(self):
        return models.Model(self.x, self.z)

    def Decoder(self):
        z_shape = (self.z_dim,)
        z = layers.Input(shape=z_shape)
        y_layer = self.layers[-1]
        y = y_layer(z)
        return models.Model(z, y)


def show_ae(autoencoder):
    encoder = autoencoder.Encoder()
    decoder = autoencoder.Decoder()

    encoded_imgs = encoder.predict(X_test)
    decoded_imgs = decoder.predict(encoded_imgs)

    n = 10
    plt.figure(figsize=(20, 6), dpi=100)
    for i in range(n):
        ax = plt.subplot(3, n, i + 1)
        plt.imshow(X_test[i].reshape(28, 28))
        plt.gray()
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)

        ax = plt.subplot(3, n, i + 1 + n)
        plt.stem(encoded_imgs[i].reshape(-1))
        plt.gray()
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)

        ax = plt.subplot(3, n, i + 1 + n + n)
        plt.imshow(decoded_imgs[i].reshape(28, 28))
        plt.gray()
        ax.get_xaxis().set_visible(False)
        ax.get_yaxis().set_visible(False)

    plt.show()


if __name__ == '__main__':
    (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:])))
 
    x_nodes = 784
    z_dim = 36

    autoencoder = AE(x_nodes, z_dim)

    history = autoencoder.fit(X_train, X_train,
                              epochs=10, batch_size=256,
                              shuffle=True, validation_data=(X_test, X_test))

    plot_history(history)

    show_ae(autoencoder)
    plt.show()