Hands-On Neuroevolution with Python

CPPN encoding

The proposed encoding scheme employs a method of representing the connectivity patterns within the phenotype ANN by querying another specialized neural network about the weights of the connections between the nodes. This specialized neural network is called a CPPN. Its main task is to represent the connectivity patterns of the phenotype ANN as a function of its geometry. The resulting connectivity pattern is represented as a four-dimensional hypercube. Each point of the hypercube encodes the connection between two related nodes within the phenotype ANN and is described by four numbers: the coordinates of the source node and the coordinates of the target node. The connective CPPN takes as input each point of the hypercube and ...