private void Breed(ref INeuralNetwork[] orderedNetworks)
{
var top = orderedNetworks.Take((orderedNetworks.Length - 1) / 2).ToArray();
var newnetwork = new List <INeuralNetwork>();
for (int i = 0; i < _speciesCount; i++)
{
var index = i % (top.Length - 1);
if (index < top.Length)
{
newnetwork.Add((INeuralNetwork)top[index].Clone());
}
}
NeuralNetworks = newnetwork;
orderedNetworks = FitnessVal.Maximize ? NeuralNetworks.OrderByDescending(p => p.Error).ToArray() : NeuralNetworks.OrderBy(p => p.Error).ToArray();
}
public IGeneration EvolveGeneration()
{
#if Roslyn
Parallel.ForEach(NeuralNetworks, p =>
{
var fitness = FitnessVal.EvaluateFitness(p);
p.Error = fitness;
});
#endif
#if Bridge
foreach (var neuralNetwork in NeuralNetworks)
{
var fitness = FitnessVal.EvaluateFitness(neuralNetwork);
neuralNetwork.Error = fitness;
}
#endif
var orderedNetworks = FitnessVal.Maximize ? NeuralNetworks.OrderByDescending(p => p.Error).ToArray() : NeuralNetworks.OrderBy(p => p.Error).ToArray();
//keep top 50
//todo: copy top 50 networks and evolve the new networks
var gen = new Generation(orderedNetworks, orderedNetworks[0], orderedNetworks[orderedNetworks.Length - 1], orderedNetworks[(orderedNetworks.Length - 1) / 2], Generation);
Breed(ref orderedNetworks);
var r = orderedNetworks.ToList();
r.Reverse();
var reverse = r.Take((int)(orderedNetworks.Length - orderedNetworks.Length * 0.1));
#if Roslyn
Parallel.ForEach(reverse, Mutate);
#endif
#if Bridge
foreach (var neuralNetwork in reverse)
{
Mutate(neuralNetwork);
}
#endif
OnGenerationFinished(gen);
Generation++;
return(gen);
}
