private void btnProcess_Click(object sender, EventArgs e)
{
NeuronNet net = new NeuronNet(new int[] { 4096, 4096, 4 });
Bitmap bmp = new Bitmap(picboxReferenceImage.Image);
float[] inputArray = new float[4096];
int i = 0;
for (int x = 0; x < bmp.Width; x++)
{
for (int y = 0; y < bmp.Height; y++)
{
inputArray[i] = bmp.GetPixel(x,y).GetBrightness();
i++;
}
}
float[] resultArray = net.Process(inputArray);
picboxLayer0.Image = (Image)net.bmps[0];
picboxLayer1.Image = (Image)net.bmps[1];
picboxLayer2.Image = (Image)net.bmps[2];
textBoxQuad.Text = resultArray[0].ToString();
textBoxCircle.Text = resultArray[1].ToString();
textBoxTriangle.Text = resultArray[2].ToString();
textBoxStar.Text = resultArray[3].ToString();
}
private void FastCrossIndividuals(NeuronNet individualA, NeuronNet individualB)
{
if (individualA.Layers.Count != individualB.Layers.Count)
{
throw new Exception("Different layer size. Can not cross");
}
for (int i = 0; i < individualA.Layers.Count; i++)
{
if (individualA.Layers[i].Neurons.Count != individualB.Layers[i].Neurons.Count)
{
throw new Exception("Different neuron size. Can not cross");
}
for (int j = 0; j < individualA.Layers[i].Neurons.Count; j++)
{
if (individualA.Layers[i].Neurons[j].Dendrites.Count != individualB.Layers[i].Neurons[j].Dendrites.Count)
{
throw new Exception("Different dendrite count. Can not cross");
}
for (int k = 0; k < individualA.Layers[i].Neurons[j].Dendrites.Count; k++)
{
double auxDendriteWeightA = individualA.Layers[i].Neurons[j].Dendrites[k].Weight;
double auxDendriteWeightB = individualB.Layers[i].Neurons[j].Dendrites[k].Weight;
individualA.Layers[i].Neurons[j].Dendrites[k].Weight = auxDendriteWeightB;
individualB.Layers[i].Neurons[j].Dendrites[k].Weight = auxDendriteWeightA;
}
}
}
}
private void DieOfAge(int maxAge)
{
if (PopulationList.Exists(x => x.Age > maxAge))
{
PopulationList = PopulationList.OrderBy(x => x.Age).ToList();
NeuronNet firstOld = PopulationList.First(x => x.Age > maxAge);
int index = PopulationList.IndexOf(firstOld);
PopulationList.RemoveRange(index, PopulationList.Count - index);
PopulationList = PopulationList.OrderBy(x => x.Error).ToList();
}
}
public void Init(GenomeFloatString genome)
{
neuronNet = new NeuronNet(new int[]{5, 8, 3}, genome.GetArray());
sensors = new float[5+1];
for (int i = 0; i < directions.Length; i++) {
directions[i].Normalize();
}
sensors[5] = 0;
totalDistance = 0;
Duration = 0;
isDone = false;
initialized = true;
}
private void learnNN(NeuronNet pNN, List <double> inputs, List <double> expected)
{
pNN.Update(inputs);
var outputs = pNN.Outputs;
Debug.Assert(outputs.Count == expected.Count);
double tempFitness = pNN.Fitness;
for (int ipOutput = 0; ipOutput < outputs.Count; ipOutput++)
{
tempFitness += calculateFitness(outputs[ipOutput], expected[ipOutput]);
}
pNN.Fitness = tempFitness;
}
public void Init(GenomeFloatString genome)
{
neuronNet = new NeuronNet(new int[] { 5, 8, 3 }, genome.GetArray());
sensors = new float[5 + 1];
for (int i = 0; i < directions.Length; i++)
{
directions[i].Normalize();
}
sensors[5] = 0;
totalDistance = 0;
Duration = 0;
isDone = false;
initialized = true;
}
private void NearGenesCrossNoOffspring(NeuronNet individualA, NeuronNet individualB)
{
if (individualA.Layers.Count != individualB.Layers.Count)
{
throw new Exception("Different layer size. Can not cross");
}
for (int i = 0; i < individualA.Layers.Count; i++)
{
if (individualA.Layers[i].Neurons.Count != individualB.Layers[i].Neurons.Count)
{
throw new Exception("Different neuron size. Can not cross");
}
for (int j = 0; j < individualA.Layers[i].Neurons.Count; j++)
{
if (individualA.Layers[i].Neurons[j].Dendrites.Count != individualB.Layers[i].Neurons[j].Dendrites.Count)
{
throw new Exception("Different dendrite count. Can not cross");
}
for (int k = 0; k < individualA.Layers[i].Neurons[j].Dendrites.Count; k++)
{
double auxDendriteWeightA = individualA.Layers[i].Neurons[j].Dendrites[k].Weight;
double auxDendriteWeightB = individualB.Layers[i].Neurons[j].Dendrites[k].Weight;
double maximum;
double minimum;
if (auxDendriteWeightA > auxDendriteWeightB)
{
maximum = auxDendriteWeightA;
minimum = auxDendriteWeightB;
}
else
{
maximum = auxDendriteWeightB;
minimum = auxDendriteWeightA;
}
individualA.Layers[i].Neurons[j].Dendrites[k].Weight = Rnd.NextDouble() * (maximum - minimum) + minimum;
individualB.Layers[i].Neurons[j].Dendrites[k].Weight = Rnd.NextDouble() * (maximum - minimum) + minimum;
}
}
}
}
static void Main(string[] args)
{
_network = new NeuronNet(2, 3, 1);
var learning = new BackPropLearning(_network);
var param = new BackPropParams()
{
CallBack = cb,
Eta = 0.9,
Alpha = 0.05,
ErrorStopValue = 0.05
};
learning.TrainNetworkBySample(XorSamples, XorSamples, param);
//learning.TrainNetworkByBatch(XorSamples, XorSamples, param);
//var param2 = new RPropParams();
//var learning2 = new RPropLearning(_network);
//learning2.Train(XorSamples, XorSamples, param2);
Console.WriteLine("press enter...");
Console.ReadLine();
}
public NaturalDescent(NeuronNet network)
{
_network = network;
}
public void TrainGenetic(int iterationsNumber, int populationSize, int[] shape)
{
_geneticAlgorithm = new GeneticAlgorithm(shape, populationSize, Rnd, -1, 1, _inputTrainSet.First());
_geneticAlgorithm.RunGenerations(iterationsNumber, _inputTrainSet, _outputTrainSet, 4);
_network = _geneticAlgorithm.PopulationList.First();
}
public void TrainGenetic(List <double[]> inputTrainSet, List <double[]> outputTarget, int iterationsNumber, int populationSize, int[] shape)
{
_geneticAlgorithm = new GeneticAlgorithm(shape, populationSize, Rnd, -1, 1, inputTrainSet[0]);
_geneticAlgorithm.RunGenerations(iterationsNumber, inputTrainSet, outputTarget, 4);
_network = _geneticAlgorithm.PopulationList.First();
}
public void CreateNetwork(int[] shape)
{
_network = new NeuronNet(shape, Rnd);
_naturalDescent = new NaturalDescent(_network);
}
private void NearGenesCrossWithOffspring(NeuronNet individualA, NeuronNet individualB)
{
NeuronNet offspring = new NeuronNet(Shape, Rnd);
double extraRange = 0.1;
offspring.SetInputLayer(InputExample);
if (individualA.Layers.Count != individualB.Layers.Count)
{
throw new Exception("Different layer size. Can not cross");
}
for (int i = 0; i < individualA.Layers.Count; i++)
{
if (individualA.Layers[i].Neurons.Count != individualB.Layers[i].Neurons.Count)
{
throw new Exception("Different neuron size. Can not cross");
}
for (int j = 0; j < individualA.Layers[i].Neurons.Count; j++)
{
double auxBiasA = individualA.Layers[i].Neurons[j].Bias;
double auxBiasB = individualB.Layers[i].Neurons[j].Bias;
double maximumBias;
double minimumBias;
if (auxBiasA > auxBiasB)
{
maximumBias = auxBiasA + (auxBiasA * extraRange);
minimumBias = auxBiasB - (auxBiasA * extraRange);
}
else
{
maximumBias = auxBiasB + (auxBiasA * extraRange);
minimumBias = auxBiasA - (auxBiasA * extraRange);
}
offspring.Layers[i].Neurons[j].Bias = Rnd.NextDouble() * (maximumBias - minimumBias) + minimumBias;
if (individualA.Layers[i].Neurons[j].Dendrites.Count != individualB.Layers[i].Neurons[j].Dendrites.Count)
{
throw new Exception("Different dendrite count. Can not cross");
}
for (int k = 0; k < individualA.Layers[i].Neurons[j].Dendrites.Count; k++)
{
double auxDendriteWeightA = individualA.Layers[i].Neurons[j].Dendrites[k].Weight;
double auxDendriteWeightB = individualB.Layers[i].Neurons[j].Dendrites[k].Weight;
double maximum;
double minimum;
if (auxDendriteWeightA > auxDendriteWeightB)
{
maximum = auxDendriteWeightA + (auxDendriteWeightA * extraRange);
minimum = auxDendriteWeightB - (auxDendriteWeightA * extraRange);
}
else
{
maximum = auxDendriteWeightB + (auxDendriteWeightA * extraRange);
minimum = auxDendriteWeightA - (auxDendriteWeightA * extraRange);
}
offspring.Layers[i].Neurons[j].Dendrites[k].Weight = Rnd.NextDouble() * (maximum - minimum) + minimum;
}
}
}
PopulationList.Add(offspring);
}
public void PaintOutput(NeuronNet network)
{
throw new NotImplementedException();
}
