/// <summary>
/// Runs each Neural Network of the population through the fitness function and assigns the fitnesses.
/// Overloaded for running on multiple threads. default is unthreaded.
/// </summary>
public void Run()
{
Dictionary <int, double> fitnesses = new Dictionary <int, double>();
double totalfitness = 0;
if (threaded)
{
Thread thread1 = new Thread(() =>
{
for (int i = 0; i < Networks.Count / 4; i++)
{
NeuralNetwork NN = Networks.ElementAt(i);
double fitness = run(NN);
totalfitness += fitness;
lock (fitnesses)
{
fitnesses.Add(i, fitness);
}
//Console.WriteLine(i + " -> " + fitness);
}
});
Thread thread2 = new Thread(() =>
{
for (int i = Networks.Count / 4; i < Networks.Count / 2; i++)
{
NeuralNetwork NN = Networks.ElementAt(i);
double fitness = run(NN);
totalfitness += fitness;
lock (fitnesses)
{
fitnesses.Add(i, fitness);
}
//Console.WriteLine(i + " -> " + fitness);
}
});
Thread thread3 = new Thread(() =>
{
for (int i = Networks.Count / 2; i < 3 * Networks.Count / 4; i++)
{
NeuralNetwork NN = Networks.ElementAt(i);
double fitness = run(NN);
totalfitness += fitness;
lock (fitnesses)
{
fitnesses.Add(i, fitness);
}
//Console.WriteLine(i + " -> " + fitness);
}
});
Thread thread4 = new Thread(() =>
{
for (int i = 3 * Networks.Count / 4; i < Networks.Count; i++)
{
NeuralNetwork NN = Networks.ElementAt(i);
double fitness = run(NN);
totalfitness += fitness;
lock (fitnesses)
{
fitnesses.Add(i, fitness);
}
//Console.WriteLine(i + " -> " + fitness);
}
});
thread1.Start();
thread2.Start();
thread3.Start();
thread4.Start();
thread1.Join();
thread2.Join();
thread3.Join();
thread4.Join();
}
else
{
int count = 0;
foreach (NeuralNetwork NN in Networks)
{
double fitness = run(NN);
totalfitness += fitness;
fitnesses.Add(count, fitness);
//Console.WriteLine(count + " -> " + fitness);
count++;
}
}
AvgPopFitness = totalfitness / Networks.Count;
BestNetwork = null;
for (int i = 0; i < Networks.Count(); i++)
{
fitnesses.TryGetValue(i, out double fitness);
if (BestNetwork == null || fitness > BestNetwork.Fitness)
{
BestNetwork = Networks.ElementAt(i);
}
Networks.ElementAt(i).Fitness = fitness;
}
//Console.WriteLine("Run: " + isConsistent());
}
/// <summary>
/// Performs speciation, adjusting fitness sums, culling, crossover, and mutation, to generate the next generation of Neural Networks
/// </summary>
public void Select()
{
speciesList = new List <Species>();
foreach (NeuralNetwork NN in Networks)
{
bool foundSpecies = false;
foreach (Species species in speciesList)
{
if (NeuralNetwork.EvolutionaryDistance(species.networks.ElementAt(0), NN, innovationNumbers) < SpeciatingThreshold)
{
species.networks.Add(NN);
foundSpecies = true;
break;
}
}
if (!foundSpecies)
{
speciesList.Add(new Species(NN));
}
}
//Console.WriteLine("speciesList Count after speciating: " + speciesList.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
foreach (Species species in speciesList)
{
species.networks.Sort((NN1, NN2) => NN2.Fitness.CompareTo(NN1.Fitness));
int kill = (int)(Math.Ceiling(species.networks.Count() * killRate));
int total = species.networks.Count();
for (int i = species.networks.Count() - 1; i > total - 1 - kill; i--)
{
species.networks.RemoveAt(i);
}
for (int i = 0; i < species.networks.Count(); i++)
{
species.adjustedFitnessSum += species.networks.ElementAt(i).Fitness / (species.networks.Count);
}
}
speciesList.RemoveAll(x => x.networks.Count < 3);
//Console.WriteLine("speciesList Count after killing: " + speciesList.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
int numSelectionBreed = (int)(.75 * Size);
speciesList.Sort((species1, species2) => species2.adjustedFitnessSum.CompareTo(species1.adjustedFitnessSum));
double sharedFitnessTotal = 0;
for (int i = 0; i < speciesList.Count / 3; i++)
{
Species species = speciesList.ElementAt(i);
sharedFitnessTotal += species.adjustedFitnessSum;
}
//Console.WriteLine("speciesList Count after adjusting fitness sums: " + speciesList.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
List <NeuralNetwork> childrenNetworks = new List <NeuralNetwork>();
Random rand = new Random();
for (int i = 0; i < speciesList.Count / 3; i++)
{
Species species = speciesList.ElementAt(i);
for (int j = 0; j < numSelectionBreed * (species.adjustedFitnessSum / sharedFitnessTotal); j++)
{
//Console.WriteLine("Pop Consistency: " + isConsistent());
NeuralNetwork NN1 = species.networks.ElementAt(rand.Next(species.networks.Count));
NeuralNetwork NN2 = species.networks.ElementAt(rand.Next(species.networks.Count));
NeuralNetwork Child = NeuralNetwork.Crossover(NN1, NN2, innovationNumbers);
Child.RandomMutation();
//Child.DisplayOutputConnections();
childrenNetworks.Add(Child);
//Console.WriteLine("Pop Consistency: " + isConsistent());
//Console.WriteLine();
}
}
//Console.WriteLine();
//Console.WriteLine("speciesList Count after selection breeding: "+ speciesList.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
//Console.WriteLine();
Networks.Sort((NN1, NN2) => NN2.Fitness.CompareTo(NN1.Fitness));
for (int i = 0; i < 5; i++)
{
childrenNetworks.Add(Networks.ElementAt(i));
}
int numRandomBreed = Networks.Count - childrenNetworks.Count;
for (int i = 0; i < numRandomBreed; i++)
{
Species randSpecies = speciesList[rand.Next(speciesList.Count)];
NeuralNetwork randNN1 = randSpecies.networks[rand.Next(randSpecies.networks.Count)];
NeuralNetwork randNN2 = randSpecies.networks[rand.Next(randSpecies.networks.Count)];
NeuralNetwork child = NeuralNetwork.Crossover(randNN1, randNN2, innovationNumbers);
child.RandomMutation();
//child.DisplayOutputConnections();
childrenNetworks.Add(child);
}
//Console.WriteLine();
//Console.WriteLine("speciesList Count after random breeding: " + speciesList.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
//Console.WriteLine();
Networks = new List <NeuralNetwork>(childrenNetworks);
//Console.WriteLine("total child networks after selection: " + childrenNetworks.Count);
//Console.WriteLine("Pop Consistency: " + isConsistent());
}
