public void SetRateMSE(float[][] inputs, float[][] answers)
{
Base.Parallel.ParallelFor(0, PopulationCount, (int i) =>
{
PopulationUnit unit = GetPopulationUnit(i);
float TotalMse = 0;
for (int testDataCounter = 0; testDataCounter < inputs.Length; testDataCounter++)
{
float[] NNOut = unit.NNRun(inputs[testDataCounter]); // получаем выходы нс
float MSE = 0;
for (int k = 0; k < NNOut.Length; k++)
{
MSE += (float)Math.Pow(answers[testDataCounter][k] - NNOut[k], 2); //подсчитываем сумму ошибок
}
MSE = MSE / NNOut.Length; // делим
TotalMse += MSE;
}
TotalMse /= answers.Length;
unit.Rate = TotalMse;
});
}
public object Clone()
{
var PU = new PopulationUnit(NN);
PU.NN = NN;
PU.Outputs = Outputs;
PU.Rate = Rate;
return(PU);
}
public int CompareTo(object obj)
{
PopulationUnit p = obj as PopulationUnit;
if (p != null)
{
return(this.Rate.CompareTo(p.Rate));
}
else
{
throw new Exception("Невозможно сравнить два объекта");
}
}
public PopulationController(FeedForwardNN nn, int PopulationCount)
{
if (PopulationCount < 2)
{
Console.ForegroundColor = ConsoleColor.Red;
Console.WriteLine("Population count must be more than 2");
Console.ForegroundColor = ConsoleColor.White;
}
this.PopulationCount = PopulationCount;
this.descriptor = nn.Descriptor;
Population = new PopulationUnit[PopulationCount];
for (int i = 0; i < PopulationCount; i++)
{
Population[i] = new PopulationUnit(nn);
}
}
public void CreateNewPopulation(float mutationChance, int mutationPower, float crossoverChance, bool bestMin = false)
{
PopulationUnit best1;
PopulationUnit best2;
Array.Sort(Population);
if (!bestMin)
{
best1 = (PopulationUnit)Population[PopulationCount - 1].Clone();
best2 = (PopulationUnit)Population[PopulationCount - 2].Clone();
}
else
{
best1 = (PopulationUnit)Population[0].Clone();
best2 = (PopulationUnit)Population[1].Clone();
}
for (int i = 2; i < PopulationCount; i++)
{
PopulationUnit Crossover = new PopulationUnit(new FeedForwardNN(descriptor));
float cross = Base.Random.NextFloat(0, 1);
float crossType = Base.Random.NextFloat(0, 1);
int kLim = 0;
if (crossType >= 0.5f && cross < crossoverChance)
{
kLim = Convert.ToInt32(Base.Random.Next(0, best1.NN.Weights.Count - 1));
}
for (int k = 0; k < best1.NN.Weights.Count; k++)
{
for (int j = 0; j < best1.NN.Weights[k].Count; j++)
{
for (int l = 0; l < best1.NN.Weights[k][j].Count; l++)
{
Crossover.NN.Weights[k][j][l] = best1.NN.Weights[k][j][l];
if (cross < crossoverChance)
{
if (crossType < 0.5f)
{
if (Base.Random.Next(0, 2) == 0)
{
Crossover.NN.Weights[k][j][l] = best2.NN.Weights[k][j][l];
}
}
else
{
if (kLim > k)
{
Crossover.NN.Weights[k][j][l] = best2.NN.Weights[k][j][l];
}
else
{
Crossover.NN.Weights[k][j][l] = best1.NN.Weights[k][j][l];
}
}
}
if (Base.Random.NextFloat(0, 1) < mutationChance)
{
Crossover.NN.Weights[k][j][l] += Base.Random.NextFloat(-mutationPower, mutationPower);
}
}
}
}
Population[i] = Crossover;
}
best1.Rate = 0;
best2.Rate = 0;
Population[0] = best1;
Population[1] = best2;
}
