public void Train(double learningRate, int epochs, double momentum, List <TrainingElement> inputs, double desiredError = 0)
{
ResetLayers();
Errors.Clear();
for (var i = 0; i < epochs; i++)
{
List <double> epochsErrors = new List <double>();
for (var j = 0; j < inputs.Count; j++)
{
var guess = ForwardPropagation(inputs[j].Input);
epochsErrors.Add(MeanSquaredError(guess, inputs[j].DesiredOutput));
//an equation for the error in the output layer, δL
var outputLayer = Layers.Last();
var sigmoidDerivative =
outputLayer.ActivationFunction.CalculateDifferential(outputLayer.WeightedSum); //∇aC=(aL−y)
outputLayer.DeltaL = guess.Subtract(inputs[j].DesiredOutput).PointwiseMultiply(sigmoidDerivative); //δL
//an equation for the error δl in terms of the error in the next layer, δl + 1
for (var k = Layers.Count - 2; k >= 0; k--)
{
Layers[k].Backpropagate(Layers[k + 1]);
}
Layers.First().UpdateLayer(inputs[j].Input, learningRate, momentum);
for (var k = 1; k < Layers.Count; k++)
{
Layers[k].UpdateLayer(Layers[k - 1].Activation, learningRate, momentum);
}
}
Errors.Add(epochsErrors.Sum() / inputs.Count);
if (Errors[i] <= desiredError)
{
break;
}
Gatherer.GatherStatistics(this);
}
}