Hands-On Machine Learning for Algorithmic Trading

How to define the architecture

We can define a simplified version of LeNet5 that omits the original final layer containing radial basis functions as follows, using the default 'valid' padding and single step strides, unless defined otherwise:

lenet5 = Sequential([            Conv2D(filters=6, kernel_size=5, activation='relu',                          input_shape=(28, 28, 1), name='CONV1'),                   AveragePooling2D(pool_size=(2, 2), strides=(1, 1),                      padding='valid', name='POOL1'),                   Conv2D(filters=16, kernel_size=(5, 5),                       activation='tanh', name='CONV2'),                   AveragePooling2D(pool_size=(2, 2), strides=(2, 2),                      name='POOL2'),                   Conv2D(filters=120, kernel_size=(5, 5),                      activation='tanh', name='CONV3'),                   Flatten(name='FLAT'),                   Dense(units=84, activation='tanh', name='FC6'), Dense(units=10, activation='softmax', ...