NeuralNetworkWithArrays Mutate(NeuralNetworkWithArrays experimental, int mutationTimes)
{
for (int i = 0; i < mutationTimes; i++)
{
int chance = Random.Range(0, 4);
switch (chance)
{
case 0:
{
experimental.weightsInputLayer[Random.Range(0, experimental.weightsInputLayer.GetLength(0)), Random.Range(0, experimental.weightsInputLayer.GetLength(1))] = Random.Range(0.99f, -0.99f);
return(experimental);
}
case 1:
{
experimental.weightsHiddneLayers[Random.Range(0, experimental.weightsHiddneLayers.GetLength(0)), Random.Range(0, experimental.weightsHiddneLayers.GetLength(1))] = Random.Range(0.99f, -0.99f);
return(experimental);
}
case 2:
{
experimental.biasesInputLayer[Random.Range(0, experimental.biasesInputLayer.Length), 0] = Random.Range(0.99f, -0.99f);
return(experimental);
}
case 3:
{
experimental.biasesHiddneLayers[Random.Range(0, experimental.biasesHiddneLayers.Length), 0] = Random.Range(0.99f, -0.99f);
return(experimental);
}
}
}
return(experimental);//без этой строчки компилятор ругается прост
}
//поскульку классы имеют прекрасное свойство создавать ссылки друг на друга, а не копироваться, существует эта функция
NeuralNetworkWithArrays NetCopy(NeuralNetworkWithArrays net)
{
NeuralNetworkWithArrays copy = new NeuralNetworkWithArrays();
copy.number = net.number;
for (int i = 0; i < net.weightsInputLayer.GetLength(0); i++)
{
for (int j = 0; j < net.weightsInputLayer.GetLength(1); j++)
{
copy.weightsInputLayer[i, j] = net.weightsInputLayer[i, j];
}
}
for (int i = 0; i < net.weightsHiddneLayers.GetLength(0); i++)
{
for (int j = 0; j < net.weightsHiddneLayers.GetLength(1); j++)
{
copy.weightsHiddneLayers[i, j] = net.weightsHiddneLayers[i, j];
}
}
for (int i = 0; i < net.biasesInputLayer.GetLength(0); i++)
{
for (int j = 0; j < net.biasesInputLayer.GetLength(1); j++)
{
copy.biasesInputLayer[i, j] = net.biasesInputLayer[i, j];
}
}
for (int i = 0; i < net.biasesHiddneLayers.GetLength(0); i++)
{
for (int j = 0; j < net.biasesHiddneLayers.GetLength(1); j++)
{
copy.biasesHiddneLayers[i, j] = net.biasesHiddneLayers[i, j];
}
}
return(copy);
}
//просто скрещенивание. берем чистое существо. подбрасывая монетку даём ему ген, то от одного, то от друго родителя
List <NeuralNetworkWithArrays> Crossover(NeuralNetworkWithArrays parent1, NeuralNetworkWithArrays parent2)//возвращаем агент с сетью
{
NeuralNetworkWithArrays childNet1;
NeuralNetworkWithArrays childNet2;
NeuralNetworkWithArrays p1Net = parent1;
NeuralNetworkWithArrays p2Net = parent2;
childNet1 = NetCopy(p1Net);
childNet2 = NetCopy(p2Net);
#region legacy
/*
* //веса c входного слоя на скрытый
* for (int i = 0; i < Random.Range(0, p1Net.weightsInputLayer.GetLength(0)); i++)
* {
* for (int j = 0; j < Random.Range(0, p1Net.weightsInputLayer.GetLength(1)); j++)
* {
* childNet1.weightsInputLayer[i, j] = p2Net.weightsInputLayer[i, j];
* }
* }
*
* for (int i = Random.Range(0, p1Net.weightsInputLayer.GetLength(0)); i < p1Net.weightsInputLayer.GetLength(0); i++)
* {
* for (int j = Random.Range(0, p1Net.weightsInputLayer.GetLength(1)); j < p1Net.weightsInputLayer.GetLength(1); j++)
* {
* childNet2.weightsInputLayer[i, j] = p1Net.weightsInputLayer[i, j];
* }
* }
*
*
* //веса со скрытого на выходной
* for (int i = 0; i < Random.Range(0, p1Net.weightsHiddneLayers.GetLength(0)); i++)
* {
* for (int j = 0; j < Random.Range(0, p1Net.weightsHiddneLayers.GetLength(1)); j++)
* {
* childNet1.weightsHiddneLayers[i, j] = p2Net.weightsHiddneLayers[i, j];
* }
* }
*
* for (int i = Random.Range(0, p1Net.weightsHiddneLayers.GetLength(0)); i < p1Net.weightsHiddneLayers.GetLength(0); i++)
* {
* for (int j = Random.Range(0, p1Net.weightsHiddneLayers.GetLength(1)); j < p1Net.weightsHiddneLayers.GetLength(1); j++)
* {
* childNet2.weightsHiddneLayers[i, j] = p1Net.weightsHiddneLayers[i, j];
* }
* }
*
*
* //веса входного слоя
* for (int i = 0; i < Random.Range(0, p1Net.biasesInputLayer.GetLength(0)); i++)
* {
* childNet1.biasesInputLayer[i, 0] = p2Net.biasesInputLayer[i, 0];
* }
*
* for (int i = Random.Range(0, p1Net.biasesInputLayer.GetLength(0)); i < p1Net.biasesInputLayer.GetLength(0); i++)
* {
* childNet2.biasesInputLayer[i, 0] = p1Net.biasesInputLayer[i, 0];
* }
*
*
* //веса скрытого слоя
* for (int i = 0; i < Random.Range(0, p1Net.biasesHiddneLayers.GetLength(0)); i++)
* {
* childNet1.biasesHiddneLayers[i, 0] = p2Net.biasesHiddneLayers[i, 0];
* }
*
* for (int i = Random.Range(0, p1Net.biasesHiddneLayers.GetLength(0)); i < p1Net.biasesHiddneLayers.GetLength(0); i++)
* {
* childNet1.biasesHiddneLayers[i, 0] = p1Net.biasesHiddneLayers[i, 0];
* }
*/
#endregion
//веса c входного слоя на скрытый
for (int i = 0; i < p1Net.weightsInputLayer.GetLength(0); i++)
{
for (int j = 0; j < p1Net.weightsInputLayer.GetLength(1); j++)
{
if (Random.value > 0.5f)
{
childNet1.weightsInputLayer[i, j] = p2Net.weightsInputLayer[i, j];
childNet2.weightsInputLayer[i, j] = p1Net.weightsInputLayer[i, j];
}
}
}
//веса со скрытого на выходной
for (int i = 0; i < Random.Range(0, p1Net.weightsHiddneLayers.GetLength(0)); i++)
{
for (int j = 0; j < Random.Range(0, p1Net.weightsHiddneLayers.GetLength(1)); j++)
{
if (Random.value > 0.5f)
{
childNet1.weightsHiddneLayers[i, j] = p2Net.weightsHiddneLayers[i, j];
childNet2.weightsHiddneLayers[i, j] = p1Net.weightsHiddneLayers[i, j];
}
}
}
//веса входного слоя
for (int i = 0; i < Random.Range(0, p1Net.biasesInputLayer.GetLength(0)); i++)
{
if (Random.value > 0.5f)
{
childNet1.biasesInputLayer[i, 0] = p2Net.biasesInputLayer[i, 0];
childNet2.biasesInputLayer[i, 0] = p1Net.biasesInputLayer[i, 0];
}
}
//веса скрытого слоя
for (int i = 0; i < Random.Range(0, p1Net.biasesHiddneLayers.GetLength(0)); i++)
{
if (Random.value > 0.5f)
{
childNet1.biasesHiddneLayers[i, 0] = p2Net.biasesHiddneLayers[i, 0];
childNet1.biasesHiddneLayers[i, 0] = p1Net.biasesHiddneLayers[i, 0];
}
}
List <NeuralNetworkWithArrays> childs = new List <NeuralNetworkWithArrays>();
childs.Add(childNet1);
childs.Add(childNet2);
return(childs);
}
