private void BackPropagate(Neuron outputNeuron)
{
for (var neuronIndex = 0; neuronIndex < HiddenLayer.Neurons.Count; neuronIndex++)
{
var neuron = HiddenLayer.Neurons[neuronIndex];
neuron.Error = Sigmoid.Derivative(neuron.Output) * outputNeuron.Error * outputNeuron.Weights[neuronIndex];
neuron.AdjustWeights();
}
}
public void Train(List <DataSet> dataSets)
{
if (!ValidateDataSets(dataSets))
{
throw new Exception("The data set is invalid for this network.");
}
var epoch = 0;
while (epoch < MaxEpochs)
{
foreach (var dataSet in dataSets)
{
//Input Initialization
foreach (var neuron in InputLayer.Neurons)
{
neuron.Inputs[0] = dataSet.Values[InputLayer.Neurons.IndexOf(neuron)];
}
//Forward Propagation Through Hidden Layer
foreach (var neuron in HiddenLayer.Neurons)
{
neuron.Inputs = InputLayer.Neurons.Select(x => x.Output).ToArray();
}
//Output Layer
foreach (var neuron in OutputLayer.Neurons)
{
neuron.Inputs = HiddenLayer.Neurons.Select(x => x.Output).ToArray();
neuron.Error = Sigmoid.Derivative(neuron.Output) * (dataSet.Results[OutputLayer.Neurons.IndexOf(neuron)] - neuron.Output);
neuron.AdjustWeights();
//Back Propagation
BackPropagate(neuron);
}
}
epoch++;
}
}
