public void AddPattern(Pattern pattern)
{
if (pattern.Input.Count != _layers.InputLayer.Neurons.Count ||
pattern.Output.Count != _layers.OutputLayer.Neurons.Count)
throw new ArgumentException("The pattern must match the number of neurons in the input and output layers");
_patterns.Add(pattern);
}
void SetInputLayer(Pattern pattern)
{
for (int i = 0; i < _layers.InputLayer.Neurons.Count; i++)
{
_layers.InputLayer.Neurons[i] = pattern.Input[i];
}
}
void PrintError(Pattern pattern)
{
double error = 0.0;
for (int num = 0; num < pattern.Output.Count; num++)
error += pattern.Output[num] - _layers.OutputLayer.Neurons[num];
error = error / pattern.Output.Count;
Console.Out.WriteLine("Error = " + error);
}
void MoveForward(Pattern pattern)
{
this.SetInputLayer(pattern);
for (int i = 0; i < _weights.Count; i++)
{
// Calculate Net.j = SUM.i W.i.j * a.i + Bias.j
double netInput = 0.0;
for (int j = 0; j < _layers[i + 1].Neurons.Count; j++)
{
for (int k = 0; k < _layers[i].Neurons.Count; k++)
{
netInput += _layers[i].Neurons[k] * _weights[i][k][j]; // +_layers[i + 1].Bias;
}
// Calculate a.j = f(Net.j)
_layers[i + 1].Neurons[j] = this.ActivationFunction(netInput);
}
}
}
void MoveBackward(Pattern pattern)
{
this.CalculateLearningDeltas(pattern);
this.BackpropagateErrors();
}
void CalculateLearningDeltas(Pattern pattern)
{
List<double> outputDelta = _deltas[_deltas.Count - 1];
for (int i = 0; i < _layers.OutputLayer.Neurons.Count; i++)
{
// Calculate error.k = t.p - a.k
double error = pattern.Output[i] - _layers.OutputLayer.Neurons[i];
// d.k = f'(Net.k) * error.k = a.k * (1 - a.k) * error.k
outputDelta[i] = this.DeltaFunction(_layers.OutputLayer.Neurons[i]) * error;
}
}
