org.knime.base.data.neural.methods

Class RProp

java.lang.Object

org.knime.base.data.neural.methods.RProp

public class RProp
extends Object

Implementation of the RProp Algorithm, as proposed by M. Riedmiller, H.Braun: 'A Direct Adaptive Method for Faster backpropagation Learning: The RPROP Algorithm', Proc. of the IEEE Intl. Conf. on Neural Networks 1993.

Author:

Nicolas Cebron, University of Konstanz

Constructor Summary

Constructors

Constructor and Description
RProp() Constructor, uses default learning rate of 0.1, increase parameter 1.2 and decrease parameter 0.5 as proposed in the paper.
RProp(double etaPlus, double etaMinus, double etaNull)

Method Summary

Modifier and Type	Method and Description
double[]	evaluate(double[] in) Evaluates input and returns output of output neurons.
double	getEtaMinus() Get negative learning rate.
double	getEtaNull() Get starting value for eta.
double	getEtaPlus() Get positive learning rate.
void	setEtaMinus(double etaMinus) Set negative learning rate.
void	setEtaNull(double etaNull) set starting value for eta.
void	setEtaPlus(double etaPlus) Set positive learning rate.
static double	sgn(double d) Method computes the sign of a double number.
void	train(MultiLayerPerceptron nn, Double[][] samples, Double[][] outputs) Train the neural network once.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

RProp

public RProp()

Constructor, uses default learning rate of 0.1, increase parameter 1.2 and decrease parameter 0.5 as proposed in the paper.

RProp

public RProp(double etaPlus,
             double etaMinus,
             double etaNull)

Parameters:

etaPlus - increase parameter

etaMinus - decrease parameter

etaNull - initial learning rate

Method Detail

train

public void train(MultiLayerPerceptron nn,
                  Double[][] samples,
                  Double[][] outputs)

Train the neural network once.

Parameters:

nn - neural net to train

samples - the samples

outputs - the desired outputs for these samples

sgn

public static double sgn(double d)

Method computes the sign of a double number.

Parameters:

d - the number

Returns:

sgn(d)

getEtaMinus

public double getEtaMinus()

Get negative learning rate.

Returns:

current negative learning rate

getEtaPlus

public double getEtaPlus()

Get positive learning rate.

Returns:

current positive learning rate

setEtaMinus

public void setEtaMinus(double etaMinus)

Set negative learning rate.

Parameters:

etaMinus - new negative learning rate

setEtaPlus

public void setEtaPlus(double etaPlus)

Set positive learning rate.

Parameters:

etaPlus - new positive learning rate

getEtaNull

public double getEtaNull()

Get starting value for eta.

Returns:

current starting value for eta

setEtaNull

public void setEtaNull(double etaNull)

set starting value for eta.

Parameters:

etaNull - new starting value

evaluate

public double[] evaluate(double[] in)

Evaluates input and returns output of output neurons.

Parameters:

in - input for the net

Returns:

output of the output neurons after having processed a forward wave through the net