/// <summary>
/// backpropagation learning algorithm
/// </summary>
/// <param name="inputs">the input</param>
/// <param name="expected_outputs">the expected output</param>
public override void Learn(float[][] inputs, float[][] expected_outputs)
{
base.Learn(inputs, expected_outputs);
float[] NeuronOutput;
float MeanSquareError2;
CurrentEpoch = 0;
// for each epoch
do
{
// currently no error
MeanSquareError = 0f;
_error_vector = new float[ANN.NumberOfOutputs];
// run through set of training data
for (int i = 0; i < InputTrainingData.Length; i++)
{
MeanSquareError2 = 0f;
NeuronOutput = ANN.Output(inputs[i]);
for (int j = 0; j < NeuronOutput.Length; j++)
{
_error_vector[j] = OutputTrainingData[i][j] - NeuronOutput[j];
// update current error
MeanSquareError2 += _error_vector[j] * _error_vector[j];
}
MeanSquareError2 /= 2f;
MeanSquareError += MeanSquareError2;
CalculateUsefulVariable(i);
// update weight
SetWeight(i);
}
CurrentEpoch++;
// we aren't dead
Pulse();
// custom check if we have to run further
if (Convergence())
{
break;
}
}while (CurrentEpoch < _maximum_of_epochs && MeanSquareError > _error_treshold);
}