public void EndLearn()
{
//рассчитываю сигмы и, конечно, обучаю выходные нейроны
//очень важно согласовать здесь все размерности массивов
//сигмы
foreach (RBFCell cell in this.HiddenCells)
{
double MinDist = double.MaxValue;
for (int i = 0; i < this.HiddenCells.Count; i++)
{
double dist = cell.Dist(HiddenCells[i].Expectations);
if ((dist < MinDist) && (dist != 0))
{
MinDist = dist;
}
}
cell.Sigma = MinDist / 2;
if (cell.Sigma == 0)
{
cell.Sigma = 1;
}
}
//а теперь градиентный спуск
Random R = new Random();
for (int i = 0; i < this.HiddenCells.Count(); i++)
{
this.OutputCells[i].InputWidth = this.HiddenCells.Count();
this.OutputCells[i].InitWeights(() => 2 * R.NextDouble() - 1);
this.OutputCells[i].Iteration = 0;
}
for (int i = this.HiddenCells.Count(); i < this.OutputCellCount; i++)
{
this.OutputCells[i].InputWidth = 0;
this.OutputCells[i].InitWeights(() => 0);
}
double error;
//Обучаем
/*
* foreach (var pair in this.TrainingSet)
* {
* do
* {
* foreach (double[] img in pair.Value)
* {
*
* double[] ResVect = new double[this.OutputCellCount];
* ResVect[pair.Key] = 1;
* this.Input = img;
* this.DoWork();
*
* for (int i = 0; i < this.OutputCellCount; i++)
* {
* double err = ResVect[i] - this.OutputCells[i].Output;
* for (int w = 0; w < this.OutputCells[i].InputWidth; w++)
* {
* this.OutputCells[i].Weights[w] += err * this.OutputCells[i].Input[w] * this.LearnSpeed;
* this.OutputCells[i].Threshoold += err * this.LearnSpeed;
* this.OutputCells[i].Iteration++;
* }
* }
* }
* //Проверяем, насколько хорошо усвоена выборка
* error = 0;
* foreach (double[] img in pair.Value)
* {
* double[] ResVect = new double[this.OutputCellCount];
* ResVect[pair.Key] = 1;
* this.Input = img;
* this.DoWork();
* double err = RBFGrid.Dist(ResVect, this.Output) / 2;
* if (err > error) error = err;
* }
* } while (error > this.MaxErrorValue);
* }
*/
//Обучаем
do
{
foreach (var pair in this.TrainingSet)
{
foreach (double[] img in pair.Value)
{
double[] ResVect = new double[this.OutputCellCount];
ResVect[pair.Key] = 1;
this.Input = img;
this.DoWork();
for (int i = 0; i < this.OutputCellCount; i++)
{
double err = ResVect[i] - this.OutputCells[i].Output;
for (int w = 0; w < this.OutputCells[i].InputWidth; w++)
{
this.OutputCells[i].Weights[w] += err * this.OutputCells[i].Input[w] * this.LearnSpeed;
this.OutputCells[i].Threshoold += err * this.LearnSpeed;
this.OutputCells[i].Iteration++;
}
}
}
}
//Проверяем, насколько хорошо усвоена выборка
error = 0;
foreach (var pair in this.TrainingSet)
{
foreach (double[] img in pair.Value)
{
double[] ResVect = new double[this.OutputCellCount];
ResVect[pair.Key] = 1;
this.Input = img;
this.DoWork();
double err = RBFGrid.Dist(ResVect, this.Output) / 2;
if (err > error)
{
error = err;
}
}
}
} while (error > this.MaxErrorValue);
this.Input = null;
}
private void button2_Click(object sender, EventArgs e)
{
if (this.recognizer == null)
{
this.recognizer = new RBFGrid();
this.recognizer.InputCellCount = height * width;
this.recognizer.OutputCellCount = 3;
}
if (this.size == 0)
{
MessageBox.Show("No image to teach ");
return;
}
int group = 0;
if (radioButton2.Checked)
{
group = 1;
}
if (radioButton3.Checked)
{
group = 2;
}
recognizer.Input = BMPTransform.BitmapToDouble(GetBitmap(draws));
recognizer.LearnImage(group);
MessageBox.Show("Success");
}