private static void BackProgagate()
{
OutLayer.CalculateErrors();
HidLayer.CalculateErrors();
HidLayer.AdjustWeights();
InLayer.AdjustWeights();
}
public static void Test()
{
Console.WriteLine("Starts to test the neural network ...");
InitializeConfusionMatrix();
var gesamtElemente = 0;
var korrekteElemente = 0;
foreach (var image in TestData)
{
gesamtElemente++;
SetNeuronValueFromImage(image.Data); //Image data wird in InputLayer eingelesen
SetDesiredValueFromLabel(image.Label); //LabelDaten werden im OutputLayer gespeichert für späteren Vergleich
FeedForward();
for (var i = 0; i < OutLayer.DesiredValues.Length; i++)
{
if (OutLayer.DesiredValues[i].CompareTo(1) != 0)
{
continue;
}
if (OutLayer.GetHighestNeuronIndex() == i)
{
korrekteElemente++;
}
}
ConfusionMatrix[(int)image.Label, OutLayer.GetHighestNeuronIndex()]++;
}
Console.WriteLine("Creating Confusion-Matrix ...");
var accuracy = ((double)korrekteElemente * (double)100 / (double)gesamtElemente);
PrintConfusionMatrix();
Console.WriteLine($"Accuracy: {accuracy} %");
}
private void isModelParametersValid()
{
if (Network[0].Type == LayerType.Custom)
{
return;
}
//the last layer in the network should be output
if (Network.Last().Param1.ToString() != OutLayer.First().Dimension)
{
throw new Exception("The output dimension of the last layer in the network must be same as label dimension!");
}
if (Network.Where(x => x.Param1 <= 0 && (x.Type != LayerType.Normalization && x.Type != LayerType.Drop)).Count() > 0)
{
throw new Exception("Layer cannot be defined with zero output dimension.!");
}
if (Network.Where(x => x.Param2 <= 0 && (x.Type == LayerType.LSTM || x.Name.StartsWith("CudaStacked"))).Count() > 0)
{
throw new Exception("Layer cannot be defined with zero output dimension.!");
}
}
private static void FeedForward()
{
InLayer.CalculateNeuronValues();
HidLayer.CalculateNeuronValues();
OutLayer.CalculateNeuronValues();
}
