public GeneticAlgorithmOld(int populationSize, int generation, GenomeOld[] population)
{
this.populationSize = populationSize;
this.generation = generation;
this.population = population;
initialize();
}
public GeneticAlgorithmOld(int populationSize, int numInputs, int numOutputs, int numLayers, int numNeuronsPerLayer)
{
this.populationSize = populationSize;
population = new GenomeOld[populationSize];
for (int currGenome = 0; currGenome < populationSize; ++currGenome) {
population[currGenome] = new GenomeOld(NeuralNet.createRandomWeights(numInputs, numOutputs, numLayers, numNeuronsPerLayer));
}
generation = 0;
initialize();
}
//Create a new population selecting and crossing genomes of the previous population.
private void createNewPopulation()
{
//Sort the population by fitness.
Array.Sort(population);
//Save the fittests genome
mostFit = new GenomeOld(population[0]);
//Create a new population.
GenomeOld[] newPopulation = new GenomeOld[populationSize];
//Save the top genomes from the previous population.
int keepBestNum = 5;
elitism(newPopulation, keepBestNum);
//Crossover until new population is full.
for (int currGenome = keepBestNum; currGenome < populationSize; ++currGenome) {
//Select two parents at random through a roulette wheel and cross them.
GenomeOld[] children = GenomeOld.singlePointCrossover(rouletteWheelSelection(), rouletteWheelSelection(), crossoverRate);
//Add first child to the population and mutate.
newPopulation[currGenome] = children[0];
newPopulation[currGenome].mutate(mutationRate);
++currGenome;
//If population is not full, add second child and mutate.
if(currGenome < populationSize) {
newPopulation[currGenome] = children[1];
newPopulation[currGenome].mutate(mutationRate);
}
}
//Save over old population.
population = newPopulation;
}
//Save the top genomes from the previous population into the new population.
private void elitism(GenomeOld[] newPopulation, int topBest)
{
string fittest = "";
for (int currBest = 0; currBest < topBest; ++currBest) {
fittest += population[currBest].fitness +" ";
newPopulation[currBest] = new GenomeOld(population[currBest]);
}
Debug.Log ("Saving these fitness values: "+fittest);
}
//Create a population from a string.
//Used in loading
public static GeneticAlgorithmOld loadPopulationFromString(string contents)
{
char[] separators1 = new char[1];
separators1[0] = separator;
string[] c = contents.Split(separators1);
char[] separators2 = new char[1];
separators2[0] = '\n';
string[] init = c[0].Split(separators2);
int populationSize = Convert.ToInt32(init[0]);
int generationNumber = Convert.ToInt32(init[1]);
GenomeOld[] population = new GenomeOld[populationSize];
for (int currGenome = 0; currGenome < populationSize; ++currGenome) {
population[currGenome] = new GenomeOld(GenomeOld.createWeightsFromString(c[1+currGenome].Split(separators2)));
}
return new GeneticAlgorithmOld(populationSize, generationNumber, population);
}
