Hands-On Neuroevolution with Python

The substrate builder function

The HyperNEAT method is built around the notion of the substrate that defines the structure of the discriminator ANN. Therefore, it is crucial to create an appropriate substrate configuration to be used during the experiment execution. The substrate creation routines are defined in the following two functions:

• The substrate builder function create_substrate creates the substrate object as follows:

def create_substrate(dim):    # Building sheet configurations of inputs and outputs    inputs = create_sheet_space(-1, 1, dim, -1)    outputs = create_sheet_space(-1, 1, dim, 0)    substrate = NEAT.Substrate( inputs, [], # hidden outputs)    substrate.m_allow_input_output_links = True    ... substrate.m_hidden_nodes_activation = \ ...