public void learnNetworkTest()
{
int dimension = 0;
List<LearningExample> examples = new List<LearningExample>();
MLPNetwork target = null;
List<int> dimensions = new List<int>();
int iterations = 1; // Tylko jedna iteracja
try
{
target = new MLPNetwork(dimension, dimensions, null, null, examples);
}
catch (NetworkDimensionException e1)
{
Assert.AreEqual(e1.BadDimension, 0);
//dimensions.Add(3);
dimensions.Add(2);
dimensions.Add(1);
dimension = 3;
target = new MLPNetwork(dimension, dimensions, null, null, examples);
try
{
target.learnNetwork(iterations, 0.5);
}
catch (ExampleListException)
{
Vector v = new Vector(3);
v[0] = 1;
v[1] = 0.8;
v[2] = 0.3;
Perceptron p0 = target.Layers[0].Perceptrons[0];
Perceptron p1 = target.Layers[0].Perceptrons[1];
Perceptron p2 = target.Layers[1].Perceptrons[0];
p0.Weights[0] = 0.3;
p0.Weights[1] = 0.5;
p0.Weights[2] = -0.2;
p1.Weights[0] = -0.5;
p1.Weights[1] = 0.1;
p1.Weights[2] = -0.2;
p2.Weights[0] = -0.4;
p2.Weights[1] = -0.8;
p2.Weights[2] = 0.1;
Vector expected = new Vector(2);
expected[1] = 0.8;
//Console.WriteLine("Iloczyn wektorowy: " + v * p0.Weights);
//Console.WriteLine(p2.Weights.ToString());
examples.Add(new LearningExample(v, expected));
//Console.WriteLine("Wyjście pierwszego perceptronu:");
//Console.WriteLine(p0.outputFunction(v).ToString());
//Console.WriteLine("Wyjście drugiego perceptronu:");
//Console.WriteLine(p1.outputFunction(v).ToString());
Vector v0 = new Vector(3);
v0[0] = 1;
v0[1] = s(0.64);
v0[2] = s(-0.48);
target.classify(examples[0]);
v0 = round(v0);
//Console.WriteLine("Wektor po zaokrągleniu: " + v0);
// Sprawdza, czy dobrze liczone są wartości wyjściowe perceptronów
Assert.AreEqual(v0.ToString(), round(target.ClassificationExamples[0].Example).ToString());
Vector v1 = new Vector(2);
v1[0] = 1;
v1[1] = s(-0.886);
v1 = round(v1);
Assert.AreEqual(v1.ToString(), round(target.ClassificationExamples[1].Example).ToString());
Assert.AreEqual(0.292, v1[1]);
Console.WriteLine("Blad: " + target.globalError());
target.learnNetwork(1, 0.5);
Console.WriteLine("Wyjścia kolejnych warstw:");
foreach (LearningExample e in target.ClassificationExamples)
{
Console.WriteLine(round(e.Example).ToString());
}
Vector d0 = new Vector(2);
d0[1] = -0.105;
Assert.AreEqual(d0.ToString(), round(target.delty[1]).ToString());
Vector d1 = new Vector(3);
d1[1] = 0.019;
d1[2] = -0.002;
Assert.AreEqual(d1.ToString(), round(target.delty[0]).ToString());
Console.WriteLine("Wagi pierwszego perceptronu po iteracji:");
Console.WriteLine(round(p0.Weights).ToString());
Console.WriteLine("Wagi drugiego perceptronu po iteracji:");
Console.WriteLine(round(p1.Weights).ToString());
Console.WriteLine("Wagi trzeciego perceptronu po iteracji:");
Console.WriteLine(round(p2.Weights).ToString());
for (int i = 0; i < 500; i++)
{
target.learnNetwork(iterations, 0.5);
Console.WriteLine("Blad: " + target.globalError());
}
return;
}
}
Assert.Fail("Nie został zrzucony odpowiedni wyjątek");
}
/// <summary>
/// Funkcja przeznaczona do tworzenia nowej sieci.
/// </summary>
private void createNetwork()
{
// Tworzy nowy obiekt i przekazuje przykłady uczące
network = new MLPNetwork(3, learningExamples);
// Przypisanie obsługi zdarzeń
network.OnNetworkCreated += networkCreated;
network.OnPerceptronCreated += perceptronCreated;
// Tworzy sieć
network.createNetwork((int)((double)Dispatcher.Invoke(getLiczbaDel)),
(double)Dispatcher.Invoke(getStalaDel));
// Wypisuje kolejne przykłady uczące i ich klasyfikacje
foreach (LearningExample l in learningExamples)
{
Dispatcher.Invoke(printFDel, l.Example.ToString() + " => " + network.classify(l) +
" (Wartość oczekiwana: " + l.ExpectedDoubleValue + ")");
}
}