/// <summary>
/// Mixs the weights of 2 Neural Networks with a chance (mutationRate) to randomise a weight.
/// Usefull when combined with Genetic Algorithms
/// </summary>
/// <returns>The newest Neural Network</returns>
/// <param name="papa">First Neural Network.</param>
/// <param name="maman">Second Neural Network.</param>
/// <param name="sizeNetwork">Size network.</param>
/// <param name="mutationRate">Mutation rate (Between 0 and 1).</param>
public static ArtificialNeuralNetwork MixNeuron(ArtificialNeuralNetwork papa, ArtificialNeuralNetwork maman, int[] sizeNetwork, float mutationRate)
{
ArtificialNeuralNetwork baby = new ArtificialNeuralNetwork(sizeNetwork);
//recopy every neuron up to the output layer
for (int iLayer = 0; iLayer < sizeNetwork.Length - 1; iLayer++)
{
//each neuron in the layer i, +1 for bias
for (int jNeuron = 0; jNeuron < sizeNetwork[iLayer] + 1; jNeuron++)
{
//each weight in the current Neuron
for (int kWeight = 0; kWeight < papa.GetNeuron(iLayer, jNeuron).weights.Length; kWeight++)
{
float weight = 0;
if ((float)Random.Range(0f, 1f) < mutationRate)
{
weight = ArtificialNeuralNetwork.Neuron.RandomWeight();
}
else
{
if (Random.Range(0, 2) == 0)
{
weight = papa.GetNeuron(iLayer, jNeuron).weights[kWeight];
}
else
{
weight = maman.GetNeuron(iLayer, jNeuron).weights[kWeight];
}
}
baby.GetNeuron(iLayer, jNeuron).weights[kWeight] = weight;
}
}
}
return(baby);
}
public static void SaveBrain(this ArtificialNeuralNetwork nn, string additional)
{
string str = "";
for (int i = 0; i < nn.SizeNetwork.Length - 1; i++)
{
//+1 for bias
for (int j = 0; j < nn.SizeNetwork[i] + 1; j++)
{
if (nn.NeuronNetwork[i, j] == null)
{
continue;
}
for (int k = 0; k < nn.NeuronNetwork[i, j].weights.Length; k++)
{
str += nn.NeuronNetwork[i, j].weights[k];
if (k != nn.NeuronNetwork[i, j].weights.Length - 1)
{
str += " ";
}
}
str += "\n"; //each neuron
}
// str += "\n"; //each layer
}
string path = "Assets/Resources/NeuralNetworkValue" + additional + ".txt";
StreamWriter writer = new StreamWriter(path, false);
writer.WriteLine(str);
writer.Close();
}
//Broken
public static void LoadBrain(this ArtificialNeuralNetwork nn, string additional)
{
string path = "Assets/Resources/NeuralNetworkValue" + additional + ".txt";
StreamReader reader = new StreamReader(path);
for (int i = 0; i < nn.SizeNetwork.Length - 1; i++)
{
//+1 for bias
for (int j = 0; j < nn.SizeNetwork[i] + 1; j++)
{
string[] strWeights = reader.ReadLine().Split(' ');
float[] weights = new float[strWeights.Length];
for (int k = 0; k < strWeights.Length; k++)
{
//Debug.Log(strWeights[k]);
weights[k] = System.Convert.ToSingle(strWeights[k]);
}
nn.NeuronNetwork[i, j].weights = weights;
}
}
reader.Close();
}
public static void SaveBrain(this ArtificialNeuralNetwork nn)
{
SaveBrain(nn, "");
}
public static void LoadBrain(this ArtificialNeuralNetwork nn)
{
LoadBrain(nn, "");
}