public void Perceptron_CanBackProp_Test()
{
int howInputs = 60;
int howOnHidden = 60;
Perceptron p = new Perceptron(
ActivationFunctionType.Sigmoid,
new int[] { howInputs, howOnHidden, 1 },
false,
0.7,
0.3);
double[] input = Enumerable.Range(1, howInputs).Select(i => (double)i).ToArray();
double[] ideal = new double[] { 0.6 };
double[] first = p.GetResult(input);
System.Console.WriteLine(p.LossFunction(ideal));
p.BackPropagation(ideal);
double[] second = p.GetResult(input);
System.Console.WriteLine(p.LossFunction(ideal));
p.BackPropagation(ideal);
double[] third = p.GetResult(input);
System.Console.WriteLine(p.LossFunction(ideal));
}
/// <summary>
/// Обучает нейросеть
/// </summary>
/// <param name="ideal">Эталонные значения</param>
/// <returns>Ошибки на первом слое</returns>
public double[,,] BackPropagation(double[] ideal)
{
var deltasOnPercInp = perceptron.BackPropagation(ideal);
var deltaOnPl1Inp = pl1.UpSample(
deltasOnPercInp.ToArray3(
pl1.MaxPositions.GetLength(0) / pl1.HowLayers,
pl1.MaxPositions.GetLength(1) / poolSize,
pl1.MaxPositions.GetLength(2) / poolSize));
var deltaOnCl2Inp = cl2.BackPropagation(deltaOnPl1Inp);
var deltaOnCl1Inp = cl1.BackPropagation(deltaOnCl2Inp);
return(deltaOnCl1Inp);
}
static void Train(Perceptron p, int times)
{
for (int i = 0; i < times; i++)
{
double[][] d = TrainingData();
p.Predict(d[0]);
p.BackPropagation(d[1], 0.45, 0.85);
double[] res = p.Result;
if ((double)i / times > 0.97)
{
Console.WriteLine($"[{d[0][0]}, {d[0][1]}] -> {res[0].ToString("0.000")} ~ {d[1][0]}");
}
}
}
