Hands-On Machine Learning for Algorithmic Trading

Defining a neural network architecture with placeholders

Keras contains a wrapper that we can use with the sklearn GridSearchCV class. It requires a build_fn instance, which constructs and compiles the model based on arguments that can later be passed during the GridSearchCV iterations.

The following make_model function illustrates how to flexibly define various architectural elements for the search process. The dense_layers argument defines both the depth and width of the network as a list of integers. We also use dropout for regularization, expressed as a float in the range [0, 1], to define the probability that a given unit will be excluded from a training iteration, as follows:

def make_model(dense_layers, activation, dropout):    '''Creates ...