public void RealizeEvolution()
{
Initialisation initialisation = new Initialisation(initialPopulationCount);
// Initialisation - validated
List <Path> population = initialisation.GenerateInitialPopulation();
for (int i = 0; i < population.Count; i++)
{
if (StaticOperations.ValidatePath(population[i]) == false)
{
throw new NotSupportedException();
}
}
// Evaluation
Evaluation evaluation = new Evaluation();
evaluation.EvaluatePopulation(population);
// Encoding
List <Representation> representations = new List <Representation>();
Decoder decoder = new Decoder();
foreach (Path path in population)
{
Representation representation = decoder.EncodePath(path);
representations.Add(representation);
}
// Evolution cycle
for (int i = 0; i < evolutionCycles; i++)
{
// Selection
Selection selection = new Selection(parentsCount, representations);
List <Representation> parents = selection.SelectParents();
// Genetic operator - validated
GeneticOperator geneticOperator = new GeneticOperator(descendantsCount, parents);
List <Representation> descendants = geneticOperator.GenerateDescendants();
// Decoding
List <Path> descendantPaths = new List <Path>();
foreach (Representation representation in descendants)
{
Path path = decoder.DecodeRepresentation(representation);
if (StaticOperations.ValidatePath(path) == false)
{
throw new NotSupportedException();
}
descendantPaths.Add(path);
}
// Evaluation
evaluation.EvaluatePopulation(descendantPaths);
for (int j = 0; j < descendants.Count; j++)
{
descendants[j].Fitness = descendantPaths[j].Fitness;
}
// Replacement
Replacement replacement = new Replacement(parents, descendants, initialPopulationCount);
representations = replacement.NextGeneration();
// Save to export file
SaveTwoBestMembers(representations);
}
}
/// <summary>
/// |C|だけ個体を生成
/// </summary>
private void GeneticOperation()
{
var Parent = new Tree[2];
Func <int, Tree> Selector = SelectByRoulette;
for (var i = 0; i < Children.Length;)
{
double p = rand.NextDouble();
Parent[0] = Selector(TournamentSize);
if (p < RateOfCross && i < Children.Length - 1)
{
// 交叉
Parent[1] = Selector(TournamentSize);
var children = GeneticOperator.Crossover(Parent, buySign);
for (var j = 0; j < children.Length; j++)
{
if (children[j].ToString() == Parent[j].ToString())
{
children[j] = GeneticOperator.Mutation(children[j], buySign);
}
}
Children[i] = children[0];
Children[i + 1] = children[1];
i += 2;
}
else
{
// 突然変異
Children[i] = GeneticOperator.Mutation(Parent[0], buySign);
i++;
}
}
}
public void RealizeEvolution()
{
Random random = new Random();
Initialisation initialisation = new Initialisation(initialPopulationCount);
// Initialisation - validated
List <Path> population = initialisation.GenerateInitialPopulation();
for (int i = 0; i < population.Count; i++)
{
if (StaticOperations.ValidatePath(population[i]) == false)
{
throw new NotSupportedException();
}
}
// Evaluation
Evaluation evaluation = new Evaluation();
evaluation.EvaluatePopulation(population);
// Encoding
List <Representation> representations = new List <Representation>();
Decoder decoder = new Decoder();
foreach (Path path in population)
{
Representation representation = decoder.EncodePath(path);
representations.Add(representation);
}
// Evolution cycle
for (int i = 0; i < evolutionCycles; i++)
{
Console.Write("Epoch #" + i + ".");
// Reinitialisation happens every 1/10th iteration and randomly resets half of population
// Elite 10% is untouched by this process
if ((i % 500) == 0 && i != 0)
{
ReinitializePopulation(representations, (int)(3 * initialPopulationCount / 4));
}
// Remap fitness using exponential remapping
//RemapFitness(representations, remapParameter);
// Selection
Selection selection = new Selection(parentsCount, representations);
List <Representation> parents = selection.SelectParents();
//List<Representation> parents = selection.SelectCombinedParents();
// Genetic operator - validated
GeneticOperator geneticOperator = new GeneticOperator(descendantsCount, parents);
List <Representation> descendants = geneticOperator.GenerateDescendants();
// Decoding
List <Path> descendantPaths = new List <Path>();
foreach (Representation representation in descendants)
{
Path path = decoder.DecodeRepresentation(representation);
if (StaticOperations.ValidatePath(path) == false)
{
throw new NotSupportedException();
}
descendantPaths.Add(path);
}
// Evaluation
evaluation.EvaluatePopulation(descendantPaths);
for (int j = 0; j < descendants.Count; j++)
{
descendants[j].Fitness = descendantPaths[j].Fitness;
}
// Revaluate current population after fitness remapping
//List<Path> currentPaths = new List<Path>();
//foreach (Representation representation in representations)
//{
// Path path = decoder.DecodeRepresentation(representation);
// currentPaths.Add(path);
//}
//evaluation.EvaluatePopulation(currentPaths);
//for (int j = 0; j < representations.Count; j++)
//{
// representations[j].Fitness = currentPaths[j].Fitness;
//}
// Replacement
Replacement replacement = new Replacement(representations, descendants, initialPopulationCount);
//representations = replacement.GenerationReplacement();
//representations = replacement.NextGeneration();
representations = replacement.DuplicationElimination(7, representations.Count / 20 < 3 ? representations.Count / 20 : 3, 20);
Console.Write(" Maximum fitness: " + representations.Max(item => item.Fitness));
// Save to export file
SaveSixBestMembers(representations);
}
}
private void NewPopulation()
{
KillAll();
var genomes = GetAllGenomes();
var newPopulation = new List <Brain>();
if (m_Save)
{
Save(m_Generation != 1, genomes);
}
var elite = GeneticOperator.Elitism(genomes);
foreach (var genome in elite)
{
for (int j = 0; j < m_GrowthActions; j++)
{
genome.Growth();
}
newPopulation.Add(CreateKartWithBrain(genome));
}
for (int i = 0; i < (m_PopulationSize - GeneticOperator.ELITISM_NUMBER) / 2; i++)
{
Genome parent1 = GeneticOperator.TournamentSelection(genomes);
Genome parent2 = GeneticOperator.TournamentSelection(genomes);
Genome parent3 = GeneticOperator.TournamentSelection(genomes);
Genome offspring1, offspring2;
GeneticOperator.BlendCrossover(parent1, parent2, out offspring1);
GeneticOperator.BlendCrossover(parent1, parent3, out offspring2);
/*
* Genome parent1 = GeneticOperator.TournamentSelection(genomes);
* Genome parent2 = GeneticOperator.TournamentSelection(genomes);
*
* Genome offspring1, offspring2;
*
* GeneticOperator.UniformCrossover(parent1, parent2, out offspring1, out offspring2);
*/
GeneticOperator.Mutate(ref offspring1);
GeneticOperator.Mutate(ref offspring2);
for (int j = 0; j < m_GrowthActions; j++)
{
offspring1.Growth();
offspring2.Growth();
}
newPopulation.Add(CreateKartWithBrain(offspring1));
newPopulation.Add(CreateKartWithBrain(offspring2));
}
for (int i = 0; i < m_PopulationSize; i++)
{
Destroy(m_Population[i].gameObject);
}
m_Population.Clear();
m_Population = newPopulation;
InitializeBrains();
m_CurrentActionNumber += m_GrowthActions;
m_Generation++;
}
