public void Train(int numberOfEpochs, DataGetter dataGetter)
{
InputData = new List <double>();
ExpectedData = new List <double>();
ErrorX = new List <double>();
ErrorY = new List <double>();
InputData = dataGetter.getInputData();
ExpectedData = dataGetter.getExpectedData();
double error = 0;
for (int i = 0; i < numberOfEpochs; i++)
{
error = 0;
/* WYBOR SASIADUJACYCH NEURONOW DLA KAZDEGO NEURONU ABY ZAKTUALIZOWAC SIGME */
foreach (var radialNeuron in HiddenLayer)
{
var tempRadialNeurons = HiddenLayer.OrderBy(n => n.EuclideanDistance(radialNeuron)).Take(5).Skip(1).ToList(); //gdyby cos nie dzialalo warto to przeanaliozwac
radialNeuron.UpdateSigma(tempRadialNeurons);
}
for (int j = 0; j < InputData.Count; j++)
{
/* FORWARD PROPAGATION - RADIAL NEURONS */
HiddenLayer.ForEach(rn => rn.GaussianFunction(InputData[j]));
/* FORWARD PROPAGATION - NEURONS */
OutputLayer.ForEach(n => n.CalculateOutputValue(HiddenLayer));
/* BACKPROPAGATION - NEURONS */
OutputLayer.ForEach(n => n.CalculateGradient(ExpectedData[j]));
OutputLayer.ForEach(n => n.UpdateWeights(LearnRate, Momentum, HiddenLayer));
error += MSE(ExpectedData[j]);
}
error = 1.0 * error / ExpectedData.Count;
ErrorX.Add(i);
ErrorY.Add(error);
//if(i % (numberOfEpochs / 10.0) == 0)
// Console.WriteLine(error);
}
Console.WriteLine("Blad sredniokwadratowy aproksymacji: " + error + "\n");
}
public List <double> Test(DataGetter dataGetter)
{
InputData = new List <double>();
ExpectedData = new List <double>();
InputData = dataGetter.getInputData();
ExpectedData = dataGetter.getExpectedData();
double error = 0;
List <double> vs = new List <double>();
for (int i = 0; i < dataGetter.getInputData().Count; i++)
{
HiddenLayer.ForEach(rn => rn.GaussianFunction(dataGetter.getInputData()[i]));
OutputLayer.ForEach(n => n.CalculateOutputValue(HiddenLayer));
vs.Add(OutputLayer[0].OutputValue);
error += MSE(ExpectedData[i]);
}
error = 1.0 * error / ExpectedData.Count;
Console.WriteLine("Blad sredniokwadratowy aproksymacji: " + error);
return(vs);
}
