public void InitializeNetwork()
{
classes = DataClass.CreateDataClasses(bow);
classes = TestFunctions.CreateFullSet(classes, bow);
myVector x = new myVector();
{
x = bow.GetVectorsList()[0];
WithoutHiddenLayerNetwork = NeuralConstruction.CreateDefaultNetwork(x.GetVector().Count, classes);
NeuralConstruction.SampleWeight(WithoutHiddenLayerNetwork, bow.GetVectorsList(), classes);
NeuralNetwork = NeuralConstruction.CreateNewDefaultNetwork(x.GetVector().Count, classes, 5);
}
}
private void _Start_Copy_Click(object sender, RoutedEventArgs e)
{
myVector x = bow.GetVectorsList()[0];
TestClass T = TestFunctions.CreateTest2(classes);
List <myVector> vectors = test.GetTestVectors();
List <int> kNNResultsIds = new List <int>();
List <int> NNResultsIds = new List <int>();
List <int> WHNNResultsIds = new List <int>();
// Liczymy
foreach (myVector V in vectors)
{
int id = 0;
id = kNN.CalculateKNN(V, classes, 3);
kNNResultsIds.Add(id);
id = NeuralConstruction.OldSampleInput(V, WithoutHiddenLayerNetwork);
WHNNResultsIds.Add(id);
id = NeuralConstruction.SampleInput(V, NeuralNetwork);
NNResultsIds.Add(id);
}
// Dodajemy wyniki dla KNN
for (int i = 0; i < kNNResultsIds.Count; i++)
{
TestResult testresult = new TestResult("kNNAlgorithm");
testresult.filltestData(T.GetTrainingClasses(), kNNResultsIds[i], vectors[i]);
kNNResults.Add(testresult);
}
// Dodajemy wyniki dla sieci neuronowej
for (int i = 0; i < NNResultsIds.Count; i++)
{
TestResult testresult = new TestResult("NNAlgorithm");
testresult.filltestData(classes, NNResultsIds[i], vectors[i]);
NNResults.Add(testresult);
}
// Dodajemy wyniki dla sieci neuronowej bez warstwy ukrytej
for (int i = 0; i < WHNNResultsIds.Count; i++)
{
TestResult testresult = new TestResult("WHNNAlgorithm");
testresult.filltestData(classes, WHNNResultsIds[i], vectors[i]);
WHNNResults.Add(testresult);
}
ListView1.ItemsSource = kNNResults;
ListView2.ItemsSource = NNResults;
ListView3.ItemsSource = WHNNResults;
kNNResults = new List <TestResult>();
NNResults = new List <TestResult>();
WHNNResults = new List <TestResult>();
test = new TestClass();
test = TestFunctions.CreateVectorTest(bow);
}
private void UczenieSieci_Click(object sender, RoutedEventArgs e)
{
myVector x = bow.GetVectorsList()[0];
BackPropagation.UczenieSieci(200, test.GetTrainingtVectors(), NeuralNetwork, classes);
Console.Beep();
foreach (myVector testvector in test.GetTestVectors())
{
int id = NeuralConstruction.SampleInput(testvector, NeuralNetwork);
var output = NeuralNetwork.getNetwork().Where(o => o.type == 2).ToList();
}
UczenieSieci.IsEnabled = false;
}
public static void UczenieSieci(int k, List <myVector> V, Network N, List <DataClass> classes)
{
float Eta = 0.1f;
List <float> ErrorLst = new List <float>();
float ErrorTotal;
float delta;
float target;
float coef = 0.0f;
var middle = N.getNetwork().Where(o => o.type == 1).ToList();
var output = N.getNetwork().Where(o => o.type == 2).ToList();
int count = k;
while (k > 0)
{
foreach (var vector in V)
{
NeuralConstruction.SampleInput(vector, N);
int id = classes.Where(o => o.GetVectors().Contains(vector)).First().GetID();
// Teaching the Output Part basing on the Hidden Part
var HiddenLayer = N.getNetwork().Where(o => o.type == 1);
foreach (var neuron in HiddenLayer)
{
var connections = neuron.GetConnections();
for (int i = 0; i < connections.Count; i++)
{
target = (id == i) ? 1 : 0;
delta = (connections[i].To.Input - target) * (connections[i].To.Input * (1 - connections[i].To.Input)) * neuron.Input;
connections[i].NewWeight = connections[i].Weight - (Eta * delta);
}
}
// Teaching the Hidden Part basing on the Input Part
// Calculating the coeficient - I made it compute in here because it is constatnt for all outputs
// And the second part of the computations
var InputLayer = N.getNetwork().Where(o => o.type == 0);
foreach (var neuron in InputLayer)
{
var connections = neuron.GetConnections();
for (int i = 0; i < connections.Count; i++)
{
var ConTo = connections[i].To.GetConnections();
coef = 0;
for (int j = 0; j < ConTo.Count; j++)
{
target = (id == j) ? 1 : 0;
coef += ConTo[j].Weight * (ConTo[j].To.Input * (1 - ConTo[j].To.Input)) * (ConTo[j].To.Input - target);
}
delta = coef * (connections[i].To.Input * (1 - connections[i].To.Input)) * neuron.Input;
connections[i].NewWeight = connections[i].Weight - (Eta * delta);
}
}
// Finalisation
AssingNewWeights(N);
}
k--;
}
}
