private void AddLayer_Click(object sender, RoutedEventArgs e)
{
if (isLayersBaseCreated == false) // first create Layers Vector, if incorrect layers number: return
{
if (Int32.TryParse((NumberOfLayersBox.Text), out _))
{
int numberOfLayers = Convert.ToInt32(NumberOfLayersBox.Text);
if (numberOfLayers <= 0)
{
MessageBox.Show("Incorrect layers number.");
return;
}
else
{
NumberOfLayersBox.IsEnabled = false;
layers = new LayerCharacteristic[numberOfLayers];
isLayersBaseCreated = true;
MainWindow.NumberOfLayers = numberOfLayers;
}
}
}
if (currentLayer < layers.Length)
{
if (ActivationFunctionComboBox.SelectionBoxItem.ToString() == "SigmoidUnipolar")
{
layers[currentLayer] = new LayerCharacteristic(Convert.ToInt32(CurrentLayerNeuronsBox.Text), new SigmoidUnipolarFunction());
}
else if (ActivationFunctionComboBox.SelectionBoxItem.ToString() == "SigmoidBipolar")
{
layers[currentLayer] = new LayerCharacteristic(Convert.ToInt32(CurrentLayerNeuronsBox.Text), new SigmoidBipolarFunction());
}
else if (ActivationFunctionComboBox.SelectionBoxItem.ToString() == "IdentityFunction")
{
layers[currentLayer] = new LayerCharacteristic(Convert.ToInt32(CurrentLayerNeuronsBox.Text), new IdentityFunction());
}
CurrentLayerNumber.Text = (currentLayer + 1).ToString();
currentLayer++;
}
else
{
MessageBox.Show("All layers already have description");
}
}
public void OutputWithoutBiasTest()
{
var layers = new LayerCharacteristic[2]
{
new LayerCharacteristic(2, new SigmoidUnipolarFunction()),
new LayerCharacteristic(1, new IdentityFunction())
};
var net = new NeuralNetwork(1, layers, false);
var input = Vector <double> .Build.Dense(1);
input[0] = 2;
net.Layers[0].Weights[0, 0] = 1.5;
net.Layers[0].Weights[0, 1] = 1 / 4.0;
net.Layers[1].Weights[0, 0] = 1 / 4.0;
net.Layers[1].Weights[1, 0] = 2;
var output = net.CalculateOutput(input);
Assert.AreEqual(1.483, Math.Round(output[0], 3));
}
static void Main(string[] args)
{
string learningFileName = @"C:\Users\Jakub\Desktop\approximation_train_1.txt";
string testFileName = @"C:\Users\Jakub\Desktop\approximation_test.txt";
LayerCharacteristic[] layers = new LayerCharacteristic[2];
layers[0] = new LayerCharacteristic(2, new SigmoidUnipolarFunction());
layers[1] = new LayerCharacteristic(1, new IdentityFunction());
var network = new NeuralNetwork(1, layers);
ILearningProvider dataProvider = new LearningApproximationDataProvider(learningFileName, testFileName, 1, 1, true);
//dataProvider.LearnSet[0] = new Datum(Vector<double>.Build.Dense(2, 1), Vector<double>.Build.Dense(2, 1));
//dataProvider.LearnSet[1] = new Datum(Vector<double>.Build.Dense(2, 1), Vector<double>.Build.Dense(2, 0));
var trainer = new OnlineTrainer(new MeanSquareErrorCalculator(), dataProvider, new BackPropagationAlgorithm(new LearningRateHandler(0.01, 0.7, 1.05, 1.04), 0.2, 1.05));
network.ConsoleDisplay();
Console.ReadLine();
#region old demo
//LayerCharacteristic[] layers = new LayerCharacteristic[2];
//layers[0] = new LayerCharacteristic(2, new SigmoidUnipolarFunction());
//layers[1] = new LayerCharacteristic(2, new IdentityFunction());
//var network = new NeuralNetwork(1, layers);
//var input = Vector<double>.Build.Dense(2);
//input[0] = 1;
//input[1] = 2;
//foreach (var layer in network.Layers)
//{
// Console.Write("\nNext layer:");
// foreach (var row in layer.Weights.ToRowArrays())
// {
// Console.WriteLine("\nrow: ");
// foreach (var cell in row)
// {
// Console.Write($"{cell} | ");
// }
// }
//}
//Console.WriteLine($"\nCalculated first: {network.CalculateOutput(input)[0]}");
//Console.WriteLine($"\nCalculated seond: {network.CalculateOutput(input)[1]}");
//var r = network.CalculateOutput(input);
//Console.ReadLine();
#endregion
#region multiplty demo
//var m = Matrix<double>.Build.Random(2,1);
//var v = Vector<double>.Build.Dense(2,2);
//Console.WriteLine($"vector: {v[0]} {v[1]}");
//Console.Write("\nMatrix:");
//foreach (var row in m.ToRowArrays())
//{
// Console.WriteLine("\nrow: ");
// foreach (var cell in row)
// {
// Console.Write($"{cell} | ");
// }
//}
//var res = v * m;
//Console.WriteLine($"res: {res.ToString()}");
//Console.ReadLine();
#endregion
}
