public void feedForward(Layer prevLayer)
{
input = 0.0;
for (int n = 0; n < prevLayer.neurons.Count(); n++)
{
input += prevLayer.neurons[n].Output * prevLayer.neurons[n].getConnection(myIndex).Weight;
}
output = MathUtils.TransferFunction(input);
}
public Net(List<int> topology)
{
int numLayers = topology.Count();
Random rand = new Random();
for (int layerNum = 0; layerNum < numLayers; layerNum++)
{
int numOutputs;
if (layerNum == numLayers - 1)
numOutputs = 0;
else
numOutputs = topology[layerNum + 1];
//Console.WriteLine("Layer number {0}", layerNum);
Layer layer = new Layer(topology[layerNum], numOutputs, rand);
m_layers.Add(layer);
}
nrInputs = topology[0];
nrOutputs = topology[numLayers - 1];
}
private double sumDOW(Layer nexLayer)
{
double sum = 0.0;
//Sum our contributions of the errors at the nodes we feed
for (int n = 0; n < nexLayer.neurons.Count() - 1; n++)
{
sum += m_outputWeights[n].Weight * nexLayer.neurons[n].m_gradient;
}
return sum;
}
public void calcHiddenGradient(Layer nextLayer)
{
double dow = sumDOW(nextLayer);
m_gradient = dow * MathUtils.TransferFunctionDerivative(input);
}
public void updateInputWeights(Layer prevLayer)
{
for (int n = 0; n < prevLayer.neurons.Count(); n++)
{
Neuron neuron = prevLayer.neurons[n];
double oldDeltaWeight = neuron.getConnection(myIndex).DeltaWeight;
//eta = overall learning rate
//alpha = momentum -> adds a fraction of the previous deltaweight
double newDeltaWeight = eta * neuron.Input * m_gradient + alpha * oldDeltaWeight;
neuron.m_outputWeights[myIndex].DeltaWeight = newDeltaWeight;
neuron.m_outputWeights[myIndex].Weight += newDeltaWeight;
}
}
