private static void MutateNeuronBiasRandomize(Ann ann)
{
ann.GetAllNeurons()[Util.rand.Next(ann.GetAllNeurons().Count)].Bias = Util.rand.NextDouble();
}
public Ann Train(int maxGenerations, int generationSize, double?errorThreshold, int mutationRate, int mutationRolls, bool waitForFlag, double[][] inputs, double[][] expectedOutputs, Crossover.CrossoverOperation crossoverOperation, ActivationFunction.ActivationMethod activationMethod)
{
Ann[] generation = new Ann[generationSize];
int inputCount = inputs[0].Length;
int outputCount = expectedOutputs[0].Length;
for (int generationId = 0; generationId < maxGenerations; generationId++)
{
if (generationId == 0)
{
for (int startingIndividual = 0; startingIndividual < generationSize; startingIndividual++)
{
Ann ann = new Ann(inputCount, outputCount, activationMethod);
double error = CalculateError(ann, inputs, expectedOutputs);
ann.Error = error;
generation[startingIndividual] = ann;
}
}
else
{
Ann parentA = generation[0];
Ann parentB = generation[1];
for (int newIndividual = 0; newIndividual < generationSize; newIndividual++)
{
generation = generation.OrderByDescending(x => x.Error).ToArray();
Ann child = crossoverOperation.Invoke(parentA, parentB);
Mutations.RollToCauseRandomMutation(child, mutationRate, mutationRolls);
double error = CalculateError(child, inputs, expectedOutputs);
child.Error = error;
if (child.Error < generation[0].Error)
{
generation[0] = child;
}
}
}
generation.ToList().ForEach(x =>
{
x.Error = CalculateError(x, inputs, expectedOutputs);
x.Generation = generationId;
});
generation = generation.OrderBy(x => x.Error).ToArray();
if (waitForFlag)
{
waitFlag = true;
OnGenerationEnd?.Invoke(generation[0]);
while (waitFlag)
{
}
}
else
{
OnGenerationEnd?.Invoke(generation[0]);
}
if (errorThreshold != null && generation[0].Error < errorThreshold)
{
return(generation[0]);
}
}
return(generation[0]);
}