Hands-On Machine Learning for Algorithmic Trading

Convolutional autoencoders

The insights from Chapter 17, Convolutional Neural Networks, suggest we incorporate convolutional layers into the autoencoder to extract information characteristic of the grid-like structure of image data.

We define a three-layer encoder that uses 2D convolutions with 32, 16, and 8 filters, respectively, ReLU activations, and 'same' padding to maintain the input size. The resulting encoding size at the third layer is 4 x 4 x 8 = 128, higher than for the preceding examples:

x = Conv2D(filters=32, kernel_size=(3, 3), activation='relu',               padding='same', name='Encoding_Conv_1')(input_)x = MaxPooling2D(pool_size=(2, 2), padding='same',                   name='Encoding_Max_1')(x)x = Conv2D(filters=16, kernel_size=(3, 3), activation='relu', ...