public Perceptron Optimize() {
Settings.Instance.UseCustomLearningRate = true;
data = TrainingSet.Load(Settings.Instance.DataDirectoryPath);
tests = TrainingSet.Load(Settings.Instance.TestsDirectoryPath);
Initialize();
Evaluation();
string info = -1 + " > best: " + population.First().Instance.Inspect() + "; " + population.First().LearningRate + "\n";
Console.WriteLine(info);
for (int i = 0; i < Settings.Instance.EvolutionEpochs; i++) {
Selection();
Crossover();
Mutation();
Evaluation();
population = population.OrderBy(c => c.Unfitness).ToArray();
info = i + " > best: " + population.First().Instance.Inspect() + "; " + population.First().LearningRate + "\n";
Console.WriteLine(info);
}
return population.First().Instance;
}
public double Learn(TrainingSet trainingSet, bool verbose = true) {
if (Settings.Instance.UseCustomLearningRate)
LearningRate = Settings.Instance.LearningRate;
else
LearningRate = 1.0 / trainingSet.Size();
double maxError = 0;
for (int step = 0; step < Settings.Instance.MaxLearningSteps; step++) {
maxError = 0;
foreach (Example example in trainingSet)
maxError = Math.Max(maxError, LearnExample(example));
if (verbose)
Engine.Output.WriteLine(step + ": " + maxError);
if (maxError <= Settings.Instance.MaxError)
break;
trainingSet.Shuffle(random);
}
return maxError;
}
public void Evaluate(TrainingSet data, TrainingSet tests) {
Instantiate();
Settings.Instance.LearningRate = LearningRate;
double dataError = Instance.Learn(data, false);
double testsError = Instance.Test(data);
const double alpha = 0.75;
Unfitness = alpha * dataError + (1.0 - alpha) * testsError;
}
public double Test(TrainingSet trainingSet) {
if (Settings.Instance.UseCustomLearningRate)
LearningRate = Settings.Instance.LearningRate;
else
LearningRate = 1.0 / trainingSet.Size();
double maxError = 0;
foreach (Example example in trainingSet)
maxError = Math.Max(maxError, TestExample(example));
return maxError;
}
public double Test(TrainingSet trainingSet)
{
if (Settings.Instance.UseCustomLearningRate)
{
LearningRate = Settings.Instance.LearningRate;
}
else
{
LearningRate = 1.0 / trainingSet.Size();
}
double maxError = 0;
foreach (Example example in trainingSet)
{
maxError = Math.Max(maxError, TestExample(example));
}
return(maxError);
}
public double Learn(TrainingSet trainingSet, bool verbose = true)
{
if (Settings.Instance.UseCustomLearningRate)
{
LearningRate = Settings.Instance.LearningRate;
}
else
{
LearningRate = 1.0 / trainingSet.Size();
}
double maxError = 0;
for (int step = 0; step < Settings.Instance.MaxLearningSteps; step++)
{
maxError = 0;
foreach (Example example in trainingSet)
{
maxError = Math.Max(maxError, LearnExample(example));
}
if (verbose)
{
Engine.Output.WriteLine(step + ": " + maxError);
}
if (maxError <= Settings.Instance.MaxError)
{
break;
}
trainingSet.Shuffle(random);
}
return(maxError);
}