void CreatePopulation(int count)
{
List <uint> Configurate = new List <uint>
{
2,
2,
1
};
Nr = new List <NeuralIndividual>();
inputs = new List <List <double> >();
outputs = new List <List <double> >();
for (int i = 0; i < count; i++)
{
inputs.Clear();
outputs.Clear();
var tmpNr = new ANeuralNetwork(Configurate, ActivationType.PositiveSygmoid, 1);
tmpNr.LoadData("..\\..\\..\\..\\PerceptronData.txt", inputs, outputs);
tmpNr.RandomInit();
//var tmpError = tmpNr.BackPropTraining(inputs, outputs, 1, 0.0001, 1, false);//Обучение сети метдом оьбратного распространения ошибки
tmpNr.IsTrained = true;
var tm = CheckAccuracy(tmpNr);
NeuralIndividual tmpNi = new NeuralIndividual(tmpNr, tm);
Nr.Add(tmpNi);
}
}
void CrossPopulation()
{
List <Tuple <NeuralIndividual, NeuralIndividual> > neuralPair = new List <Tuple <NeuralIndividual, NeuralIndividual> >();
List <uint> Configurate = new List <uint>
{
2,
2,
1
};
ANeuralNetwork ann = new ANeuralNetwork(Configurate, ActivationType.PositiveSygmoid, 1);
ann.RandomInit();
int tmpCount = Nr.Count;
List <int> usedIndex = new List <int>();
for (int i = 0; i < tmpCount; i++)
{
int rand = new Random().Next(tmpCount);
while (usedIndex.Contains(rand))
{
rand = new Random().Next(tmpCount);
}
usedIndex.Add(rand);
if (rand == tmpCount - 1)
{
for (var layerIdx = 0; layerIdx < Nr[i].network.Weights.Count; layerIdx++)
{
for (var fromIdx = 0; fromIdx < Nr[i].network.Weights[layerIdx].Count; fromIdx++)
{
for (var toIdx = 0; toIdx < Nr[i].network.Weights[layerIdx][fromIdx].Count; toIdx++)
{
var tmpRand = new Random().Next(1000);
if (tmpRand >= 50)
{
ann.Weights[layerIdx][fromIdx][toIdx] = Nr[rand].network.Weights[layerIdx][fromIdx][toIdx];
}
else
{
ann.Weights[layerIdx][fromIdx][toIdx] = Nr[0].network.Weights[layerIdx][fromIdx][toIdx];
}
}
}
}
}
else
{
for (var layerIdx = 0; layerIdx < Nr[i].network.Weights.Count; layerIdx++)
{
for (var fromIdx = 0; fromIdx < Nr[i].network.Weights[layerIdx].Count; fromIdx++)
{
for (var toIdx = 0; toIdx < Nr[i].network.Weights[layerIdx][fromIdx].Count; toIdx++)
{
var tmpRand = new Random().Next(1000);
if (tmpRand >= 50)
{
ann.Weights[layerIdx][fromIdx][toIdx] = Nr[rand].network.Weights[layerIdx][fromIdx][toIdx];
}
else
{
ann.Weights[layerIdx][fromIdx][toIdx] = Nr[rand + 1].network.Weights[layerIdx][fromIdx][toIdx];
}
}
}
}
}
ann.IsTrained = true;
var err = CheckAccuracy(ann);
Nr.Add(new NeuralIndividual(ann, err));
}
for (int i = 0; i < Nr.Count - 1; i++)
{
for (int j = 0; j < Nr.Count - i - 1; j++)
{
if (Nr[j].error > Nr[j + 1].error)
{
var temp = Nr[j];
Nr[j] = Nr[j + 1];
Nr[j + 1] = temp;
}
}
}
Nr.RemoveRange(tmpCount, tmpCount);
}