public Perceptron(int batching, double learningRate, ErrorFunction.ErrorFunction errorFunction)
{
if (batching < 1)
{
throw new ArgumentException("Parameter batching must be a positive integer.");
}
if (learningRate <= 0)
{
throw new ArgumentException("Parameter learningRate must be a positive real number.");
}
this.Batching = batching;
this.LearningRate = learningRate;
this.Layers = new List <Layer>();
this.ErrorFunction = errorFunction;
}
internal void BackPropagate(double error, double[] labels, ErrorFunction.ErrorFunction lossFunction)
{
for (int j = 0; j < CostDerivatives.Length; j++)
{
CostDerivatives[j] = lossFunction.GetDerivativeValue(labels[j], Activations[j]);
//2.0 * (this.Layers[l].Activations[j] - labels[j]);
}
for (int j = 0; j < Activations.Length; j++)
{
BiasVectorChangeRecord[j] += ActivationFunction.GetDerivativeValue(WeightedSum[j]) * CostDerivatives[j];
for (int k = 0; k < WeightMatrix.GetLength(1); k++)
{
WeightMatrixChangeRecord[j, k] += Previous.Activations[k]
* ActivationFunction.GetDerivativeValue(WeightedSum[j])
* CostDerivatives[j];
}
}
}
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);
}
