private void button1_Click(object sender, EventArgs e)
{
double dou = 0;
listBox1.Items.Clear();
for (int i = 0; i < 10; i++)
{
//Коэффициент обучение сети
n = Convert.ToDouble(textBox_n.Text);
//Объем обучающей выборки
countRow = Convert.ToInt32(textBox_X.Text);
numHidden = Convert.ToInt32(textBox_C.Text);
countIter = Convert.ToInt32(textBox1.Text);
// чтение входных данных
x = ReaderWriter.ReadMatrix(filename, countRow, numInput);
double[] d = ReaderWriter.ReadVector(filename, countRow, numInput);
D = new double[countRow, numOutput];
RecD(d);
//Инициализация центроид
centroids = ReaderWriter.ReadMatrixC(filename, countRow, numInput, numHidden, d);
rn = new RBFNetwork(countRow, numInput, numHidden, numOutput, x, centroids, n, temp, er);
rn.countIter = countIter;
//Инициализация весов
rn.SetWeight();
// Обучение сети
rn.TrainNetwork(x, centroids, countRow, numInput, numHidden, numOutput, D);
er = rn.errTrain;
countRow = 30;
// чтение входных данных
x = ReaderWriter.ReadMatrix("iris_test.txt", countRow, numInput);
d = ReaderWriter.ReadVector("iris_test.txt", countRow, numInput);
D = new double[countRow, numOutput];
RecD(d);
// Тестирование сети
errTest = rn.TestNetwork(x, D, countRow, numInput, numOutput, numHidden);
dou += errTest;
listBox1.Items.Add(errTest);
}
listBox1.Items.Add("Result: " + dou / 10);
}
private void button_Sample_Click(object sender, EventArgs e)
{
label_err.Text = "Learning error: ";
label_Test.Text = "Testing error: ";
n = Convert.ToDouble(textBox_n.Text);
//Объем обучающей выборки
countRow = Convert.ToInt32(textBox_X.Text);
numHidden = Convert.ToInt32(textBox_C.Text);
countIter = Convert.ToInt32(textBox1.Text);
x = ReaderWriter.ReadMatrix(filename, countRow, numInput);
double[] d = ReaderWriter.ReadVector(filename, countRow, numInput);
D = new double[countRow, numOutput];
RecD(d);
//Ин центроид
centroids = ReaderWriter.ReadMatrixC(filename, countRow, numInput, numHidden, d);
Console.WriteLine(centroids);
rn = new RBFNetwork(countRow, numInput, numHidden, numOutput, x, centroids, n, temp, er);
// Console.WriteLine(rn);
rn.countIter = countIter;
//Ин весов
rn.SetWeight();
rn.TrainNetwork(x, centroids, countRow, numInput, numHidden, numOutput, D);
// Console.WriteLine(rn);
er = rn.errTrain;
label_err.Text = "Learning error: " + er;
temp = rn.temp;
// button_Sample.Enabled = false;
button_test.Enabled = true;
}
public RBFNetwork TrainNetwork(Matrix x, Matrix centroids, int countRow, int numInput, int numHidden, int numOutput, double[,] d)
{
R = centroids.Row - 1;
List <double> sko = new List <double>();
double e = 0;
double eSum = 0;
double[] error = new double[countRow]; //105
double[,] newX = new double[countRow, numOutput];
double[,] newC = new double[centroids.Row, centroids.Column];
double[,] newSi = new double[centroids.Row, centroids.Column];
double m = numOutput;
double p = countRow;
bool flg = true;
double sum = 0;
double[,] gradient = new double[numHidden, numOutput];
int t = 0;
//k-means
SetCentroidsWin(x, centroids);
Sigmoida(centroids);
double[,] u = new double[countRow, numHidden]; //3
double[,] f = new double[countRow, numHidden]; //3
Matrix y = new Matrix(countRow, numOutput); //3
double temp1 = 0;
int r = countIter;
while (r != 0)
{
r--;
eSum = 0;
//Обучение сети
for (int ti = 0; ti < x.Row; ti++)
{
for (int q = 0; q < y.Column; q++)
{
e = 0;
for (int i = 0; i < centroids.Row; i++) //3
{
sum = 0;
for (int j = 0; j < centroids.Column; j++) //4
{
sum += Math.Pow(x.Get(ti, j) - centroids.Get(i, j), 2) / (Math.Pow(sigmoida[i, j], 2));
//newC[i,j] = x.Get(ti, j) - centroids.Get(i, j)/ (Math.Pow(sigmoida[i, j], 2));
//newSi[i, j] = x.Get(ti, j) - centroids.Get(i, j) / (Math.Pow(sigmoida[i, j], 3));
}
u[ti, i] = sum;
f[ti, i] = Math.Exp(-0.5 * u[ti, i]);
temp1 = 0;
}
for (int w = 0; w < weight.Column; w++)
{
temp1 += weight.Get(q, w) * f[ti, q];
}
//Console.WriteLine(temp1);
//выход
y.Set(ti, q, temp1);
//разность
newX[ti, q] = y.Get(ti, q) - d[ti, q];
//for (int i = 0; i < centroids.Row; i++) //3
//{
// for (int j = 0; j < centroids.Column; j++) //4
// {
// centroids.Set(i, j, centroids.Get(i, j) - n * newX[ti, q] * f[ti, q] * weight.Get(i, q) * (x.Get(ti, j) - centroids.Get(i, j) / (Math.Pow(sigmoida[i, j], 2))));
// // sigmoida[i, j] = sigmoida[i, j] - n * newX[ti, q] * f[ti, q] * weight.Get(i, q) * ((x.Get(ti, j) - centroids.Get(i, j) / (Math.Pow(sigmoida[i, j], 3))));
// }
//}
for (int k = 0; k < weight.Column; k++)
{
gradient[k, q] = newX[ti, q] * f[ti, q];
//пересчитываем вес
weight.Set(k, q, weight.Get(k, q) - n * gradient[k, q]);
}
//Console.WriteLine(weight);
e += Math.Pow(y.Get(ti, q) - d[ti, q], 2);
}
eSum += e;
}
sko.Add(Math.Sqrt((eSum / (m * p - 1))));
//Console.WriteLine(sko.ElementAt(sko.Count - 1));
t++;
}
NOutputRBF(y);
for (int y1 = 0; y1 < y.Row; y1++)
{
for (int y2 = 0; y2 < y.Column; y2++)
{
Console.Write(y.Get(y1, y2) + " ");
}
Console.WriteLine(t);
}
for (int y2 = 0; y2 < sko.Count(); y2++)
{
errPr = sko.ElementAt(y2);
Console.WriteLine(errPr);
}
Console.WriteLine(sko.ElementAt(sko.Count - 1));
temp = t;
errTrain = Math.Round(errPr, 10);
RBFNetwork nr = new RBFNetwork(countRow, numInput, numHidden, numOutput, x, centroids, n, temp, errTrain);
return(nr);
}
