public float[] FeedForward(float[] inputs)
{
for (int i = 0; i < inputs.Length; i++)
{
this.Neurons[0][i] = inputs[i];
}
for (int i = 1; i < this.LayerCount; i++)
{
for (int j = 0; j < this.Neurons[i].Length; j++)
{
float value = 0f;
for (int k = 0; k < this.Neurons[i - 1].Length; k++)
{
value += this.Weights[i - 1][j][k] * this.Neurons[i - 1][k];
}
this.Neurons[i][j] = NeuroHelper.Sigmoid(value);
}
}
this.FeedForwardFinished?.Invoke(this, new FeedForwardFinishedEventArgs(this.Outputs));
return(this.Outputs);
}
private void InitWeightsAndBiases(float[][][] presetWeights = null, float[][] presetBiases = null)
{
List <float[][]> weights = new List <float[][]>();
List <float[]> biases = new List <float[]>();
for (int i = 1; i < this.layers.Length; i++)
{
List <float[]> layerWeights = new List <float[]>();
float[] layerBiases = new float[this.layers[i]];
for (int j = 0; j < this.layers[i]; j++)
{
float[] neuronWeights = new float[this.layers[i - 1]];
if (presetBiases != null && presetBiases.Length >= i && presetBiases[i - 1].Length >= j)
{
layerBiases[j] = presetBiases[i - 1][j];
}
else
{
layerBiases[j] = NeuroHelper.RandomNext();
}
for (int k = 0; k < neuronWeights.Length; k++)
{
if (presetWeights != null && presetWeights.Length >= i && presetWeights[i - 1].Length >= j && presetWeights[i - 1][j].Length >= k)
{
neuronWeights[k] = presetWeights[i - 1][j][k];
}
else
{
neuronWeights[k] = NeuroHelper.RandomNext();
}
}
layerWeights.Add(neuronWeights);
}
biases.Add(layerBiases);
weights.Add(layerWeights.ToArray());
}
this.Weights = weights.ToArray();
this.Biases = biases.ToArray();
}
public void Mutate()
{
for (int i = 1; i < this.LayerCount; i++)
{
if (NeuroHelper.RandomNext(0f, 1f) <= NeuroHelper.LayerMutationChance)
{
List <float[]> neurons = new List <float[]>();
for (int l = 0; l < this.layers.Length; l++)
{
float[] layer = new float[this.layers[l] + 1];
for (int j = 0; j < this.layers[l]; j++)
{
layer[j] = 0f;
}
neurons.Add(layer);
}
this.Neurons = neurons.ToArray();
this.layers[i]++; // increase the neuroncount of the layer
this.InitWeightsAndBiases(this.Weights, this.Biases);
}
if (NeuroHelper.RandomNext(0f, 1f) <= NeuroHelper.NeuronWeightMutationChance)
{
int neuronIndex = NeuroHelper.RandomNext(0, this.Neurons[i].Length);
int weightIndex = NeuroHelper.RandomNext(0, this.Weights[i - 1][neuronIndex].Length);
if (weightIndex < this.Weights.Length)
{
this.Weights[i - 1][neuronIndex][weightIndex] += NeuroHelper.RandomNext(-NeuroHelper.NeuronWeightMutationDefaultValue, NeuroHelper.NeuronWeightMutationDefaultValue);
}
}
if (NeuroHelper.RandomNext(0f, 1f) <= NeuroHelper.NeuronBiasMutationChance)
{
int neuronIndex = NeuroHelper.RandomNext(0, this.Neurons[i].Length);
this.Biases[i - 1][neuronIndex] += NeuroHelper.RandomNext(-NeuroHelper.NeuronBiasMutationDefaultValue, NeuroHelper.NeuronBiasMutationDefaultValue);
}
}
}
public static void Selection(Dictionary <NeuroNet, int> scoredNeuroNets)
{
IEnumerable <KeyValuePair <NeuroNet, int> > sortedBrains = scoredNeuroNets.OrderBy(x => x.Value);
for (int i = 0; i < sortedBrains.Count(); i++)
{
if (i <= sortedBrains.Count() / 2 && NeuroHelper.RandomNext(0, sortedBrains.Count() * 2) < ((sortedBrains.Count() * 2 - 2) - i * 2))
{
if (scoredNeuroNets.Count >= 2)
{
scoredNeuroNets.Remove(sortedBrains.ElementAt(i).Key);
}
}
if (i > sortedBrains.Count() / 2 && NeuroHelper.RandomNext(0, sortedBrains.Count() * 2) < (sortedBrains.Count() * 2) - i * 2)
{
if (scoredNeuroNets.Count >= 2)
{
scoredNeuroNets.Remove(sortedBrains.ElementAt(i).Key);
}
}
}
}
public static NeuroNet Crossover(NeuroNet mother, NeuroNet father)
{
NeuroNet child = InitChild(mother, father);
for (int i = 0; i < child.Weights.Length; i++)
{
for (int j = 0; j < child.Weights[i].Length; j++)
{
for (int k = 0; k < child.Weights[i][j].Length; k++)
{
if (!(k >= mother.Weights[i][j].Length && mother.Weights[i][j].Length < father.Weights[i][j].Length) &&
((k >= father.Weights[i][j].Length && mother.Weights[i][j].Length > father.Weights[i][j].Length) || NeuroHelper.RandomNext(0, 100) > 50))
{
child.Weights[i][j][k] = mother.Weights[i][j][k];
}
else
{
child.Weights[i][j][k] = father.Weights[i][j][k];
}
}
if (!(j >= mother.Biases[i].Length && mother.Biases[i].Length < father.Biases[i].Length) &&
((j >= father.Biases[i].Length && mother.Biases[i].Length > father.Biases[i].Length) || NeuroHelper.RandomNext(0, 100) > 50))
{
child.Biases[i][j] = mother.Biases[i][j];
}
else
{
child.Biases[i][j] = father.Biases[i][j];
}
}
}
return(child);
}