/// <summary>
/// Perform one training iteration.
/// </summary>
public void Iteration()
{
// Reset the BMU and begin this iteration.
_bmuUtil.Reset();
var won = new int[_outputNeuronCount];
var leastRepresentedActivation = double.PositiveInfinity;
double[] leastRepresented = null;
// Reset the correction matrix for this synapse and iteration.
//*this.correctionMatrix.clear();
// Determine the BMU for each training element.
foreach (var input in _training)
{
var bmu = _bmuUtil.CalculateBMU(input.Input);
won[bmu]++;
// If we are to force a winner each time, then track how many
// times each output neuron becomes the BMU (winner).
if (ForceWinner)
{
// Get the "output" from the network for this pattern. This
// gets the activation level of the BMU.
var output = Compute(_network, input.Input);
// Track which training entry produces the least BMU. This
// pattern is the least represented by the network.
if (output[bmu] < leastRepresentedActivation)
{
leastRepresentedActivation = output[bmu];
leastRepresented = input.Input;
}
}
Train(bmu, _network.Weights, input.Input);
if (ForceWinner)
{
// force any non-winning neurons to share the burden somewhat\
if (!ForceWinners(_network.Weights, won,
leastRepresented))
{
ApplyCorrection();
}
}
else
{
ApplyCorrection();
}
}
// update the error
_error = _bmuUtil.WorstDistance / 100.0;
}
/// <summary>
/// Calculate the error for the specified data set. The error is the largest distance.
/// </summary>
/// <param name="data">The data set to check.</param>
/// <returns>The error.</returns>
public double CalculateError(IMLDataSet data)
{
var bmu = new BestMatchingUnit(this);
bmu.Reset();
// Determine the BMU for each training element.
foreach (IMLDataPair pair in data)
{
IMLData input = pair.Input;
bmu.CalculateBMU(input);
}
// update the error
return(bmu.WorstDistance / 100.0);
}