public Perceptron Layer(int size, ActivationFunction.ActivationFunction activationFunction)
{
if (Layers.Count == 0)
{
Layer layer = new Layer(activationFunction, size);
this.Layers.Add(layer);
}
else
{
Layer lastLayer = Layers[Layers.Count - 1];
Layer layer = new Layer(activationFunction, size);
this.Layers.Add(layer);
lastLayer.Next = layer;
layer.Previous = lastLayer;
layer.BuildWeightMatrix();
}
return(this);
}
public Layer(ActivationFunction.ActivationFunction activationFunction, int size)
{
this.ActivationFunction = activationFunction;
this.Size = size;
BiasVector = new double[size];
Random rand = new Random();
double prob = rand.NextDouble();
for (int i = 0; i < BiasVector.Length; i++)
{
if (prob <= 0.5)
{
BiasVector[i] = -rand.NextDouble();
}
else
{
BiasVector[i] = rand.NextDouble();
}
}
BiasVectorChangeRecord = new double[size];
Activations = new double[size];
WeightedSum = new double[size];
CostDerivatives = new double[size];
}
public static Perceptron Train(DataSet.DataSet dataSet, int batching, int epochs, double learningRate, int hiddenLayers, int hiddenLayersSize, ActivationFunction.ActivationFunction activationFunction, ErrorFunction.ErrorFunction errorFunction)
{
Perceptron p = new Perceptron(batching, learningRate, errorFunction);
p.Layer(dataSet.FeatureSize, activationFunction);
for (int i = 0; i < hiddenLayers; i++)
{
p.Layer(hiddenLayersSize, activationFunction);
}
p.Layer(dataSet.LabelSize, activationFunction);
p.Train2(dataSet, epochs);
return(p);
}