backpropagation
Class Backpropagation

java.lang.Object
  |
  +--backpropagation.Backpropagation

public class Backpropagation

extends java.lang.Object

Implementation of the well-known backpropagation algorithm with optional number of layers.

Author: Richard Szabo
Last modification: 02/Mar/2001

Field Summary

double	absoluteError Summed absolute error on the components of the output vector.
protected double	calmingRate Calming rate of the learning rate: 1.0 means constant learning rate.
protected double	elasticity Elasticity of the sigmoid function.
protected double	elasticityRate Changing rate of the elasticity.
protected backpropagation.Neuron[][]	layers Layers of the network.
protected double	learningRate Learning rate of the network.
protected double	momentum Constant determining the influence of the previous weight change on the actual.
protected int	numberOfLayers Number of the layers, must be at least 2.
protected int[]	numberOfNeurons Number of the neurons in layers, must be at least 1.
protected double	theta Threshold of the sigmoid function.

Constructor Summary

protected	Backpropagation() Hidden constructor of the class to be used in the loadNeuro static method.
	Backpropagation(int numberOfLayers, int[] numberOfNeurons) Basic constructor.
	Backpropagation(int numberOfLayers, int[] numberOfNeurons, double learningRate, double theta, double elasticity, double calmingRate, double momentum, double elasticityRate) Creation of a new BackProp network with lots of parameters.

Method Summary

void	alterWeights() Alteration of the network with Gaussian.
protected double	d1sigmoid(double x) First derivative of the sigmoid function.
double	getCalmingRate() Returns the calming rate.
double	getElasticity() Returns the actual elasticity.
double	getElasticityRate() Returns the elasticity rate.
double	getLearningRate() Returns the actual learning rate.
double	getMomentum() Returns the momentum.
double	getTheta() Returns the theta.
double	hidden(int i, int j) Returns the output value of the jth hidden neuron in the ith layer.
protected void	init() Initialization of the network with random weights.
void	learn(double[] input, double[] output) Teaches the network according to the input-output value pair.
static Backpropagation	loadNeuro(java.lang.String path, java.lang.String name) Loads a previously learned network.
double	output(int i) Returns network output of a given neuron.
void	propagate(double[] input) Propagates the network input to get the output.
void	saveNeuro(java.lang.String path, java.lang.String name) Saves the learned network weights.
void	setCalmingRate(double calmingRate) Sets the calming rate.
void	setElasticity(double elasticity) Sets the elasticity.
void	setElasticityRate(double elasticityRate) Sets the elasticity rate.
void	setLearningRate(double learningRate) Sets the learning rate.
void	setMomentum(double momentum) Sets the momentum.
void	setTheta(double theta) Sets the theta.
protected double	sigmoid(double x) Sigmoid function working in the (-1,1) interval.
java.lang.String	toString() String representation of the network.

Methods inherited from class java.lang.Object

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

Field Detail

learningRate

protected double learningRate

Learning rate of the network.

theta

protected double theta

Threshold of the sigmoid function.

elasticity

protected double elasticity

Elasticity of the sigmoid function.

calmingRate

protected double calmingRate

Calming rate of the learning rate: 1.0 means constant learning rate.

momentum

protected double momentum

Constant determining the influence of the previous weight change on the actual. 0.0 means no influence.

elasticityRate

protected double elasticityRate

Changing rate of the elasticity. It must be in the [0.0,1.0].

numberOfLayers

protected int numberOfLayers

Number of the layers, must be at least 2.

numberOfNeurons

protected int[] numberOfNeurons

Number of the neurons in layers, must be at least 1.

layers

protected backpropagation.Neuron[][] layers

Layers of the network.

absoluteError

public double absoluteError

Summed absolute error on the components of the output vector.

Constructor Detail

Backpropagation

public Backpropagation(int numberOfLayers,
                       int[] numberOfNeurons)

Basic constructor.

Backpropagation

public Backpropagation(int numberOfLayers,
                       int[] numberOfNeurons,
                       double learningRate,
                       double theta,
                       double elasticity,
                       double calmingRate,
                       double momentum,
                       double elasticityRate)

Creation of a new BackProp network with lots of parameters.

Backpropagation

protected Backpropagation()

Hidden constructor of the class to be used in the loadNeuro static method.

Method Detail

propagate

public void propagate(double[] input)

Propagates the network input to get the output. Output is in the (-1,1) interval.

learn

public void learn(double[] input,
                  double[] output)

Teaches the network according to the input-output value pair. Output is in the (-1,1) interval.

alterWeights

public void alterWeights()

Alteration of the network with Gaussian.

saveNeuro

public void saveNeuro(java.lang.String path,
                      java.lang.String name)

Saves the learned network weights.

loadNeuro

public static Backpropagation loadNeuro(java.lang.String path,
                                        java.lang.String name)

Loads a previously learned network.

toString

public java.lang.String toString()

String representation of the network.

Overrides:

toString in class java.lang.Object

output

public double output(int i)

Returns network output of a given neuron.

hidden

public double hidden(int i,
                     int j)

Returns the output value of the jth hidden neuron in the ith layer. This function is necessary for the example of Mitchell.

getLearningRate

public double getLearningRate()

Returns the actual learning rate.

setLearningRate

public void setLearningRate(double learningRate)

Sets the learning rate.

getTheta

public double getTheta()

Returns the theta.

setTheta

public void setTheta(double theta)

Sets the theta.

getElasticity

public double getElasticity()

Returns the actual elasticity.

setElasticity

public void setElasticity(double elasticity)

Sets the elasticity.

getCalmingRate

public double getCalmingRate()

Returns the calming rate.

setCalmingRate

public void setCalmingRate(double calmingRate)

Sets the calming rate.

getMomentum

public double getMomentum()

Returns the momentum.

setMomentum

public void setMomentum(double momentum)

Sets the momentum.

getElasticityRate

public double getElasticityRate()

Returns the elasticity rate.

setElasticityRate

public void setElasticityRate(double elasticityRate)

Sets the elasticity rate.

sigmoid

protected double sigmoid(double x)

Sigmoid function working in the (-1,1) interval.

d1sigmoid

protected double d1sigmoid(double x)

First derivative of the sigmoid function.

init

protected void init()

Initialization of the network with random weights.