NEATOptions

Options & Configuration · AiDotNet.NeuralNetworks.Options

Configuration options for the NEAT neural network.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → NEATOptions

How It Works

Defaults are the parameters of Stanley & Miikkulainen (2002), Section 4.1, where the paper gives them. The weight perturbation power and cap, which the paper leaves open, follow NEAT-Python.

Constructors

Constructor Summary
NEATOptions()

Properties

Property Summary
double AddConnectionRate { get; set; } Gets or sets the probability of an add-connection mutation.
double AddNodeRate { get; set; } Gets or sets the probability of an add-node mutation.
double CompatibilityThreshold { get; set; } Gets or sets the compatibility threshold delta_t that separates species.
double CrossoverRate { get; set; } Gets or sets the fraction of offspring produced by crossover.
double DisableInheritedRate { get; set; } Gets or sets the chance that an inherited gene is disabled when it was disabled in either parent.
double DisjointCoefficient { get; set; } Gets or sets c2, the weight of disjoint genes in the compatibility distance.
double ExcessCoefficient { get; set; } Gets or sets c1, the weight of excess genes in the compatibility distance.
double InterspeciesMatingRate { get; set; } Gets or sets the interspecies mating rate.
double SigmoidSlope { get; set; } Gets or sets the slope of the steepened sigmoid 1 / (1 + e^(-slope x)).
double SurvivalThreshold { get; set; } Gets or sets the fraction of each species (best first) allowed to reproduce; the rest are eliminated.
double WeightCap { get; set; } Gets or sets the bound on connection weight magnitude.
double WeightDifferenceCoefficient { get; set; } Gets or sets c3, the weight of the average weight difference of matching genes.
double WeightMutationRate { get; set; } Gets or sets the probability that a genome’s connection weights are mutated.
double WeightPerturbPower { get; set; } Gets or sets the half-width of the uniform weight perturbation.
double WeightReplaceRate { get; set; } Gets or sets the probability that a mutated weight is replaced by a new random value instead of being perturbed.
int AddConnectionTries { get; set; } Gets or sets how many node pairs an add-connection mutation tries before giving up.
int ChampionSpeciesSize { get; set; } Gets or sets the species size above which the species champion is copied unchanged into the next generation.
int GenerationsPerTrain { get; set; } Gets or sets how many generations one supervised training step evolves.
int GenomeSizeNormalizationThreshold { get; set; } Gets or sets the genome size below which N (the normalizer of E and D) is 1.
int PopulationStagnationLimit { get; set; } Gets or sets the number of generations without population-wide improvement after which only the top two species reproduce.
int SpeciesStagnationLimit { get; set; } Gets or sets the number of generations without improvement after which a species may no longer reproduce.

Methods

Method Summary
void Validate() Throws when a value on this instance cannot produce a working model.