Hands-On Neuroevolution with Python

Evolving connective CPPNs and the HyperNEAT algorithm

The method is called HyperNEAT because it uses a modified NEAT to evolve CPPNs that represent spatial patterns in the hyperspace. Each expressed point of the pattern, which is bounded by a hypercube, represents a connection between two nodes in the lower-dimensional graph (substrate). Thus, the dimensionality of the hyperspace is twice as big as the dimensionality of the underlying lower-dimensional graph. Later in Chapter 8, ES-HyperNEAT and the Retina Problem, we will look at some examples that use two-dimensional connectivity patterns.

The HyperNEAT algorithm can be seen in the following diagram:

The general form of the HyperNEAT algorithm

Any connection gene or node gene that's added ...