Hands-On Machine Learning for Algorithmic Trading

Defining the discriminator network

Both the discriminator and generator use a deep CNN architecture, wrapped in a function:

def build_discriminator():    model = Sequential([        Conv2D(32, kernel_size=3, strides=2, input_shape=img_shape, padding='same'),        LeakyReLU(alpha=0.2),        Dropout(0.25),        Conv2D(64, kernel_size=3, strides=2, padding='same'),        ZeroPadding2D(padding=((0, 1), (0, 1))),        BatchNormalization(momentum=0.8),        LeakyReLU(alpha=0.2),        Dropout(0.25),        Conv2D(128, kernel_size=3, strides=2, padding='same'),        BatchNormalization(momentum=0.8),        LeakyReLU(alpha=0.2),        Dropout(0.25),        Conv2D(256, kernel_size=3, strides=1, padding='same'),        BatchNormalization(momentum=0.8),        LeakyReLU(alpha=0.2),        Dropout(0.25),        Flatten(),        Dense(1, activation='sigmoid') ]) ...