public SmartGenAlgorithm(GeneticAlgorithm.GeneticAlgorithm geneticAlgorithm,
NeuralNetwork.NeuralNetwork neuralNetwork)
{
_geneticAlgorithm = geneticAlgorithm;
_neuralNetwork = neuralNetwork;
IsPaused = false;
}
private void runBtn_Click(object sender, RoutedEventArgs e)
{
_populationSize = (int)populField.Value;
_citiesAmount = (short)citySlider.Value;
_crossoverProb = crossSlider.Value;
_mautationProb = mutSlider.Value;
_ga = new GA(_populationSize, _citiesAmount, canvas.Width, canvas.Height, _selectOperator, _crossoverProb, _mautationProb);
_bestSolutionSoFar = _ga.GetBestSolutionSoFar();
Thread thread = new Thread(Run);
thread.Start();
}
public static GeneticAlgorithm.GeneticAlgorithm CreateGeneticAlgorithm(int genomeSize)
{
var populationSize = Settings.Default.PopulationSize;
var geneticAlgorithm = new GeneticAlgorithm.GeneticAlgorithm(populationSize, genomeSize)
{
Selection = SelectionExtension.GetSelection(Settings.Default.Selection, populationSize,
Settings.Default.SelectionSize),
Crossover = CrossoverExtension.GetCrossover(Settings.Default.Crossover, populationSize,
Settings.Default.CrossoverProbability),
Mutation = MutationExtension.GetMutation(Settings.Default.Mutation,
Settings.Default.MutationProbability)
};
return(geneticAlgorithm);
}
public void Test2([Values(500)] int iterations, [Values(100)] int columns)
{
var testDir = Path.GetDirectoryName(Path.GetDirectoryName(TestContext.CurrentContext.TestDirectory));
var dataPath = Path.Combine(testDir ?? throw new NullReferenceException(), "..", "Data", "data.csv");
var mapper = new CsvMapper();
var data = mapper.ReadDataFromFile(dataPath, 1);
var correlation = Correlation.GetCorrelation(data);
data = data.RemoveLeastRelevantColumn(correlation, columns);
var dataSet = data.SplitData(1, 2, 0);
var trainingData = dataSet[DataType.Training];
var testingData = dataSet[DataType.Testing];
var neuralNetwork = new NeuralNetwork.NeuralNetwork {
MinWeight = -5, MaxWeight = 5
};
var activationFunction = new SigmoidFunction(-1);
neuralNetwork.ActivationFunction = activationFunction;
neuralNetwork.AddLayer(new Layer(10, columns));
neuralNetwork.AddLayer(new Layer(1, 10));
var genetic = new GeneticAlgorithm.GeneticAlgorithm(100, neuralNetwork.GetConnectionCount())
{
Selection = new TournamentSelection(100, 10),
Crossover = new TwoPointCrossover(0.65, 100),
Mutation = new FlipBitMutation(0.05)
};
var keepLooping = true;
for (var it = 0; it < iterations && keepLooping; it++)
{
foreach (var chromosome in genetic.Population)
{
neuralNetwork.SetWeights(chromosome.Genome);
Parallel.For(0, trainingData.Attributes.Count, i =>
{
var res = neuralNetwork.GetResult(trainingData.Attributes[i]);
lock (chromosome)
{
chromosome.Fitness += Math.Abs(res[0] - trainingData.ObjectClass[i]);
}
});
}
keepLooping = !genetic.Population.Any(chromosome => chromosome.Fitness < 5.0);
Console.WriteLine($@"iteration: {it} | " +
$@"best: {genetic.Population.MinBy(chromosome => chromosome.Fitness).First().Fitness}");
if (keepLooping && it < iterations - 1)
{
genetic.NextGeneration();
}
}
var best = genetic.Population.MinBy(chromosome => chromosome.Fitness).First();
neuralNetwork.SetWeights(best.Genome);
var correct = testingData.Attributes
.AsParallel()
.Where((t, i) => Math.Abs(neuralNetwork.GetResult(t)[0] - testingData.ObjectClass[i]) < 0.3)
.Count();
Console.WriteLine(
$@"Population: {100}, iterations: {iterations} - fitness: {best.Fitness:0.0} " +
$@"| correctly recognized: {correct}/{testingData.Attributes.Count}");
}
public void Test1([Values(100, 200)] int populationSize,
[Values(10, 50, 250, 1000)] int iterations,
[Values(30)] int dataSize)
{
var neuralNetwork = new NeuralNetwork.NeuralNetwork {
MinWeight = -5, MaxWeight = 5
};
var activationFunction = new TanHFunction();
neuralNetwork.ActivationFunction = activationFunction;
neuralNetwork.AddLayer(new Layer(4, 2));
neuralNetwork.AddLayer(new Layer(1, 4));
var genetic = new GeneticAlgorithm.GeneticAlgorithm(populationSize, neuralNetwork.GetConnectionCount())
{
Selection = new TournamentSelection(populationSize, populationSize / 10),
Crossover = new TwoPointCrossover(0.65, populationSize),
Mutation = new FlipBitMutation(0.05)
};
var data = new List <SampleData>();
for (var i = 0; i < dataSize; i++)
{
data.Add(new SampleData());
}
for (var it = 0; it < iterations; it++)
{
foreach (var chromosome in genetic.Population)
{
neuralNetwork.SetWeights(chromosome.Genome);
foreach (var sampleData in data)
{
var res = neuralNetwork.GetResult(sampleData.Data);
lock (genetic)
{
chromosome.Fitness += Math.Abs(res[0] - sampleData.ObjectClass);
}
}
}
if (it < iterations - 1)
{
genetic.NextGeneration();
}
}
var best = genetic.Population.MinBy(chromosome => chromosome.Fitness).First();
neuralNetwork.SetWeights(best.Genome);
var correct = data.Count(d => Math.Abs(neuralNetwork.GetResult(d.Data)[0] - d.ObjectClass) < 0.3);
Console.WriteLine(
$@"Population: {populationSize}, iterations: {iterations} - fitness: {best.Fitness:0.0} " +
$@"| correctly recognized: {correct}");
foreach (var d in data)
{
var results = neuralNetwork.GetResult(d.Data)[0];
Console.WriteLine(
$@"[{d.Data[0]: 0.00;-0.00} {d.Data[1]: 0.00;-0.00}] - class: {d.ObjectClass: 0;-0} " +
$@"| recognized: {results: 0.00;-0.00} " +
$@"| is correct: {Math.Abs(results - d.ObjectClass) < 0.3}");
}
}
