Hands-On Neuroevolution with Python

Indirect encoding of ANNs with CPPNs

In the previous chapters, you learned about the direct encoding of ANNs using the nature-inspired conception of a genotype that is mapped to the phenotype in a 1:1 ratio to represent the ANN topology. This mapping allows us to use advanced NEAT algorithm features such as an innovation number, which allows us to track when a particular mutation was introduced during the evolution. Each gene in the genome has a specific value of the innovation number, allowing fast and accurate crossover of parent genomes to produce offspring. While this feature introduces immense benefits and also reduces the computational costs needed to match the parent genomes during the recombination, the direct encoding used to encode ...