public static double Calculate(this HRBFNeuronet neuronet, double[] x)
{
var hiddenLayerOutput = neuronet.HiddenLayer.Calculate(x);
var h = hiddenLayerOutput.Zip(neuronet.HiddenLayer.Neurons, (t1, t2) => t1 * t2.W).ToList();
var output = neuronet.HiddenLayer.CalculateWeightedSum(hiddenLayerOutput);
return(output);
}
public void test2()
{
var hiddenNeuronCount = 50;
var inputNeuronCount = 2;
var data = MeteoData.Read();
var normalizedData = DataPreparated.NormalizedData(data);
data = normalizedData.Data;
var allDataSets = DataPreparated.GetLearningSets(data, inputNeuronCount);
var learningDataSize = (int)(allDataSets.Count * 0.95);
var testDataSize = allDataSets.Count - learningDataSize;
Pair <double[], double>[] learningData = new Pair <double[], double> [learningDataSize];
Pair <double[], double>[] testData = new Pair <double[], double> [testDataSize];
Array.Copy(allDataSets.ToArray(), learningData, learningDataSize);
Array.Copy(allDataSets.ToArray(), learningDataSize, testData, 0, testDataSize);
var centers = DataPreparated.ExtractCenters(data, inputNeuronCount, hiddenNeuronCount);
var nn = new HRBFNeuronet(hiddenNeuronCount, inputNeuronCount, centers.Select(t1 => t1.Item1).ToList());
for (int i = 0; i < 6; i++)
{
var l1 = new Pair <double[], double> [31];
Array.Copy(learningData, i * 31, l1, 0, 31);
nn.Learning(l1.ToList(), 2500, 0.00002);
}
var d = learningData.Select(o => o.Item2).ToList();
var y = testData.Select(o => nn.Calculate(o.Item1)).ToList();
chart1.Series.Add("y");
chart1.Series.Add("d");
chart1.Series["y"].ChartType = System.Windows.Forms.DataVisualization.Charting.SeriesChartType.Line;
chart1.Series["d"].ChartType = System.Windows.Forms.DataVisualization.Charting.SeriesChartType.Line;
for (int i = 1; i < y.Count; i++)
{
chart1.Series["y"].Points.AddXY(i, y[i]);
chart1.Series["d"].Points.AddXY(i, d[i]);
}
}
private static double OneLearningIteration(HRBFNeuronet neuronet, double[] xSet, double d, double y, double learningCoef)
{
var error = y - d;
var newW = neuronet.HiddenLayer.RecalculatedW(error, xSet, learningCoef, neuronet.HiddenLayer.ActivationW);
var newC = neuronet.HiddenLayer.RecalculatedC(error, xSet, learningCoef);
var newQ = neuronet.HiddenLayer.RecalculatedQ(error, xSet, learningCoef);
neuronet.HiddenLayer.SetParams(newW, newC, newQ);
return(error);
}
public static List <double> Learning(this HRBFNeuronet neuronet, List <Pair <double[], double> > learningSet, int epochCount, double learningCoef)
{
var errors = new List <double>();
for (int epochInd = 0; epochInd < epochCount; epochInd++)
{
var xSet = learningSet[0].Item1.ToList();
for (int learningSetInd = 0; learningSetInd < learningSet.Count; learningSetInd++)
{
var y = neuronet.Calculate(xSet.ToArray());
var error = OneLearningIteration(neuronet, xSet.ToArray(), learningSet[learningSetInd].Item2, y, learningCoef);
errors.Add(error);
xSet.RemoveAt(0);
xSet.Add(y);
}
}
return(errors);
}
public void test2()
{
var hiddenNeuronCount = 5;
var inputNeuronCount = 5;
var data = MeteoData.Read();
data = DataPreparated.NormalizedData(data).Data;
var testdata = DataPreparated.GetLearningSets(data, inputNeuronCount);
var centers = DataPreparated.ExtractCenters(data, inputNeuronCount, hiddenNeuronCount);
var nn = new HRBFNeuronet(hiddenNeuronCount, inputNeuronCount, centers.Select(t1 => t1.Item1).ToList());
nn.Learning(testdata, 200, 0.02);
var d = testdata.Select(o => o.Item2).ToList();
var y = testdata.Select(o => nn.Calculate(o.Item1)).ToList();
var l = nn.Calculate(testdata[200].Item1);
}