Hands-On Neuroevolution with Python

Direct genome encoding

Direct genome encoding attempts were used in neuroevolution methods to create ANNs that were related to neural networks with a fixed topology; that is, the network topology was determined solely by the experimenter. Here, genetic encoding (genotype) is implemented as a vector of real numbers, representing the strength (weights) of the connections between the network nodes.

The evolutionary operators modify the values of the weights vector with the mutation operator and combine the vectors of the parent organisms with the recombination (crossover) operator to produce offspring. While allowing evolutionary operators to be applied with ease, the described encoding method has some significant drawbacks. One of its main ...