INeuralNetwork.LearnBack

Syntax

LearnBack(Var InputValues: Array; Var outputValues: Array);

Parameters

InputValues. Real array of input values.

outputValues. Real array of point-of-reference output values.

Description

The LearnBack method performs iteration on learning back-propagation network.

Comments

One iteration is called an epoch of learning. Short algorithm:

Output values are calculated using current values of synapse weights.
Calculated output values are compared to the point-of-reference output values. Value of the error is defined.
The value of the error is used to calculate the delta value, which is used to correct synapse weights. See INeuralNetwork.DeltasMinimumReachedBP and INeuralNetwork.GetMaximumWeightDelta.
Synapse weights are corrected.

Iterations continue the specified number of times or until the assigned accuracy of the network has been reached, that is, till delta is small enough.

Example

As an example, a function is provided that learns the network. A back-propagation network is fed at the input of the function: the Net parameter. To execute the example, add links to the NN, IO system assemblies.

Function m_BPTrain(Net: NeuralNetwork): NeuralNetwork;
Var
    epoch, sampleNumber, i: Integer;
    NumberOfOut, NumberOfInp: Integer;
    inputs, outputs: Array Of Double;
    NetFile: File;
    TextW: ITextWriter;
Begin
    NetFile := New File.Create;
    TextW := NetFile.OpenTextWriter("C:/BPTrain.txt", True);
    NumberOfOut := Net.GetNumberOfOutputs;
    outputs := New Double[NumberOfOut];
    NumberOfInp := Net.GetNumberOfInputs;
    inputs := New Double[NumberOfInp];
    For epoch := 1 To 3000 Do
        For sampleNumber := 1 To NumberOfOut Do
        // set input values
            For i := 0 To NumberOfInp - 1 Do
                inputs[i] := 0.5;
            End For;
        // set output values
            For i := 0 To NumberOfOut - 1 Do
                outputs[i] := 0.5;
            End For;
        // set learning parameters
            Net.SetNju(0.1);
            Net.SetMju(0);
        // learn the network
            Net.LearnBack(inputs, outputs);
            TextW.WriteLnString("Epoch of learning: '" + epoch.ToString + "'");
            TextW.WriteLnString("  maximum value delta: " + net.GetMaximumWeightDelta.ToString);
            TextW.WriteLnString("  output data: ");
            outputs := Net.GetOutputValues;
            For i := 0 To NumberOfOut - 1 Do
                TextW.WriteLnString("  " + outputs[i].ToString);
            End For;
        // verify whether network has been learned
            If Net.DeltasMinimumReachedBP(0.0001) Then
                Return Net;
            End If
        End For;
    End For;
    Return Net;
End Function m_BPTrain;

After executing the example the network is learned according to the assigned parameters. Changing output data and delta value are written to the file C:/BPTrain.txt.