NEAT (NeuroEvolution of Augmenting Topologies) is an evolutionary algorithm that creates artificial neural networks. In the current implementation of NEAT-Python, a population of individual genomes is maintained. Each genome contains two sets of genes that describe how to build an artificial neural network:
Node genes, each of which specifies a single neuron. Connection genes, each of which specifies a single connection between neurons. To evolve a solution to a problem, the user must provide a fitness function which computes a single real number indicating the quality of an individual genome: better ability to solve the problem means a higher score. The algorithm progresses through a user-specified number of generations, with each generation being produced by reproduction and mutation of the most fit individuals of the previous generation.
The reproduction and mutation operations may add nodes and/or connections to genomes, so as the algorithm proceeds genomes (and the neural networks they produce) may become more and more complex. When the preset number of generations is reached, or when at least one individual (for a fitness criterion function of max; others are configurable) exceeds the user-specified fitness threshold, the algorithm terminates.