public float[] getRealOtputForCorrectionGraph(int start,
int amount, ResultForOneNeuron dataToCount, ref float[,] points)
{
float[] destination = new float[amount];
for (int i = start - 1, index = 0; i < (start + amount - 1); i++, index++)
{
destination[index] = points[i, 0] * dataToCount.w1 +
points[i, 1] * dataToCount.w2 +
points[i, 2] * dataToCount.w3 -
dataToCount.t;
}
return(destination);
}
public ResultOfTrainingSPL findWeightsAndThreshold(ref float[,] trainDataSet,
float errorMax, int amountOfPoints)
{
float w1, w2, w3, t1;
float w4, w5, w6, t2;
float w7, w8, w9, t3;
float errorS = 0.0f;
float sumError = 0.0f;
float sumErrorPrev = 0.0f;
Random rnd = new Random();
w1 = (float)rnd.NextDouble();
w2 = (float)rnd.NextDouble();
w3 = (float)rnd.NextDouble();
t1 = (float)rnd.NextDouble();
w4 = (float)rnd.NextDouble();
w5 = (float)rnd.NextDouble();
w6 = (float)rnd.NextDouble();
t2 = (float)rnd.NextDouble();
w7 = (float)rnd.NextDouble();
w8 = (float)rnd.NextDouble();
w9 = (float)rnd.NextDouble();
t3 = (float)rnd.NextDouble();
//Debug.WriteLine("Wagi dla x: {0}, {1}, {2}", w1, w2, w3);
float sub;
do
{
float x;
float y;
float z;
float desireValueX, newValueX;
float desireValueY, newValueY;
float desireValueZ, newValueZ;
for (int i = 0; i < amountOfPoints; i += 1)
{
desireValueX = trainDataSet[i + 1, 0];
desireValueY = trainDataSet[i + 1, 1];
desireValueZ = trainDataSet[i + 1, 2];
x = trainDataSet[i, 0];
y = trainDataSet[i, 1];
z = trainDataSet[i, 2];
float learningRate = (float)(1.00 / (1.00 + Math.Pow(x, 2) + Math.Pow(y, 2) + Math.Pow(z, 2)));
newValueX = x * w1 + y * w2 + z * w3 - t1;
newValueY = x * w4 + y * w5 + z * w6 - t2;
newValueZ = x * w7 + y * w8 + z * w9 - t3;
w1 = w1 - learningRate * x * (newValueX - desireValueX);
w2 = w2 - learningRate * y * (newValueX - desireValueX);
w3 = w3 - learningRate * z * (newValueX - desireValueX);
t1 = t1 + learningRate * (newValueX - desireValueX);
w4 = w4 - learningRate * x * (newValueY - desireValueY);
w5 = w5 - learningRate * y * (newValueY - desireValueY);
w6 = w6 - learningRate * z * (newValueY - desireValueY);
t2 = t2 + learningRate * (newValueY - desireValueY);
w7 = w7 - learningRate * x * (newValueZ - desireValueZ);
w8 = w8 - learningRate * y * (newValueZ - desireValueZ);
w9 = w9 - learningRate * z * (newValueZ - desireValueZ);
t3 = t3 + learningRate * (newValueZ - desireValueZ);
errorS = (float)(Math.Pow(newValueX - desireValueX, 2) +
Math.Pow(newValueY - desireValueY, 2) +
Math.Pow(newValueZ - desireValueZ, 2));
sumError += errorS;
}
//Debug.WriteLine("Wagi dla x: {0}, {1}, {2}", w1, w2, w3);
Debug.WriteLine((sumError / 2) - (sumErrorPrev / 2));
sub = Math.Abs((sumError / 2) - (sumErrorPrev / 2));
sumErrorPrev = sumError / 2;
sumError = 0;
}while (sub > errorMax);
Debug.WriteLine(sumError / 2);
ResultForOneNeuron result1
= new ResultForOneNeuron(w1, w2, w3, t1);
ResultForOneNeuron result2
= new ResultForOneNeuron(w4, w5, w6, t2);
ResultForOneNeuron result3
= new ResultForOneNeuron(w7, w8, w9, t3);
ResultOfTrainingSPL resultAll = new ResultOfTrainingSPL(result1, result2, result3);
return(resultAll);
}
