private void applyGenomesInBrains()
{
for (int i = 0; i < genomes.Length; i++)
{
brains[i] = new NeuroNetwork(neuro_input_nodes, neurons_per_layer, neuro_output_nodes, genomes[i]);
}
}
public void createGeneration()
{
if (current_generation == 0)
{
// Generate random genomes on first generation
brains = new NeuroNetwork[genomes_count];
genomes = new sbyte[genomes_count][];
fitness = new double[genomes_count];
for (int i = 0; i < brains.Length; i++)
{
genomes[i] = new sbyte[NeuroNetwork.calculateGenomeLength(neuro_input_nodes, neurons_per_layer, neuro_output_nodes)];
for (int g = 0; g < genomes[i].Length; g++)
{
genomes[i][g] = (sbyte)random.Next(-128, 127);
}
}
}
else
{
// Generate new solutions trying to improve them
improveGeneration();
}
applyGenomesInBrains();
current_generation++;
}
private NeuroNetwork[] crossover(NeuroNetwork genome1, NeuroNetwork genome2)
{
NeuroNetwork newGenome1 = new NeuroNetwork(neuro_input_nodes, neurons_per_layer, neuro_output_nodes, random),
newGenome2 = new NeuroNetwork(neuro_input_nodes, neurons_per_layer, neuro_output_nodes, random);
object[] genome1_perceptrons = genome1.crossOverSplit(), genome2_perceptrons = genome2.crossOverSplit();
newGenome1.setPerceptrons((Perceptron[])genome1_perceptrons[0], (Perceptron[])genome2_perceptrons[1]);
newGenome2.setPerceptrons((Perceptron[])genome2_perceptrons[0], (Perceptron[])genome1_perceptrons[1]);
return(new NeuroNetwork[] { newGenome1, newGenome2 });
}
private void improveGeneration()
{
// Selecting better solutions
Dictionary <NeuroNetwork, double> genomes_dictionary = new Dictionary <NeuroNetwork, double>();
for (int i = 0; i < genomes_count; i++)
{
genomes_dictionary.Add(genomes[i], fitness[i]);
}
genomes_dictionary = genomes_dictionary.OrderByDescending(key => key.Value).Take(solution_selection_for_crossover).ToDictionary(pair => pair.Key, pair => pair.Value);
// Create new genomes array
NeuroNetwork[] newGenomes = new NeuroNetwork[genomes_count];
// Keep original 2 better solutions
newGenomes[0] = genomes_dictionary.Keys.First();
newGenomes[1] = genomes_dictionary.Keys.Last();
// Cross-over solutions
NeuroNetwork[] hybrids = crossover(newGenomes[0], newGenomes[1]);
for (int i = 0; i < hybrids.Length; i++)
{
newGenomes[i + 2] = hybrids[i]; // Add crossed over solutions to new generation
}
// Mutated solutions
for (int i = hybrids.Length + 2; i < newGenomes.Length; i++)
{
int selection = random.Next(0, 4);
Perceptron[] old_ps = newGenomes[selection].getPerceptrons();
Perceptron[] new_ps = new Perceptron[old_ps.Length];
for (int o = 0; o < new_ps.Length; o++)
{
new_ps[o] = new Perceptron(old_ps[o].getWeights(), old_ps[o].getBias());
}
newGenomes[i] = new NeuroNetwork(neuro_input_nodes, neurons_per_layer, neuro_output_nodes, random, new_ps);
newGenomes[i].applyMutations(random);
}
// Replace old generation with new one
genomes = newGenomes;
}
public void createGeneration()
{
if (current_generation == 0)
{
// Generate random genomes on first generation
genomes = new NeuroNetwork[genomes_count];
fitness = new double[genomes_count];
for (int i = 0; i < genomes.Length; i++)
{
genomes[i] = new NeuroNetwork(neuro_input_nodes, neurons_per_layer, neuro_output_nodes, random);
}
}
else
{
// Generate new solutions trying to improve them
improveGeneration();
}
current_generation++;
}
