private void TrainIndividual(XorIndividual individual)
{
for (int i = 0; i < _populationParameters.TrainIterations; i++)
{
individual.Genome.Network.Train(_truthTable, TrainSpeed, TrainMoment);
}
}
private void MutateIndividual(XorIndividual individual, Random random)
{
var chance = random.NextDouble();
if (chance < _mutationParameters.IndividualMutationChance)
{
individual.Genome.MutateStructure();
}
}
public int Search(int maxIteration, double treshhold)
{
Parallel.ForEach(_population, individual => TrainIndividual(individual));
Random rand = new Random();
int iteration = 0;
while (iteration < maxIteration)
{
Parallel.ForEach(_population, individual => individual.EvaluateError(_truthTable));
_population.Sort();
BestIndividual = _population[0];
Console.WriteLine($"I: {iteration} -> {BestIndividual}");
if (BestIndividual.Error <= treshhold)
{
break;
}
var archiveChilds = _population
.Take(_populationParameters.ArchiveSize)
.ToList();
var bestChilds = new List <XorIndividual>(_populationParameters.BestChildCount);
bestChilds.AddRange(
Enumerable.Range(1, _populationParameters.BestChildCount)
.AsParallel()
.Select(i => new XorIndividual(NeatGenome.Crossover(BestIndividual.Genome, _population[i].Genome)))
.ToList()
);
var randomChildCount = _populationParameters.PopulationSize - (_populationParameters.BestChildCount + _populationParameters.ArchiveSize);
var randomChilds = Enumerable.Range(0, randomChildCount)
.AsParallel()
.Select(_ =>
{
XorIndividual randomChild = null;
var individual1 = _population[rand.Next(_population.Count)];
var individual2 = _population[rand.Next(_population.Count)];
if (individual1 == individual2)
{
randomChild = individual1;
//randomChild.Genome.MutateStructure();
}
else
{
randomChild = new XorIndividual(NeatGenome.Crossover(individual1.Genome, individual2.Genome));
}
return(randomChild);
})
.ToList();
_population = new List <XorIndividual>(_populationParameters.PopulationSize);
_population.AddRange(bestChilds);
_population.AddRange(randomChilds);
_population.AddRange(archiveChilds);
// Apply mutations
Parallel.ForEach(bestChilds, individual => MutateIndividual(individual, rand));
Parallel.ForEach(randomChilds, individual => MutateIndividual(individual, rand));
// Train population
Parallel.ForEach(_population, individual => TrainIndividual(individual));
iteration++;
}
return(iteration);
}
