keel.Algorithms.Neural_Networks.EvRBF_CL

Class Rbfn

java.lang.Object

keel.Algorithms.Neural_Networks.EvRBF_CL.Rbfn

All Implemented Interfaces:

java.lang.Cloneable

public class Rbfn
extends java.lang.Object
implements java.lang.Cloneable

Class representing a Radial Basis Function Neural Network for the EvRBF_CL algorithm

Since:

JDK1.5

Version:

0.1

Author:

Written by Antonio Jesus Rivera Rivas (University of Jaen) 03/03/2004, Modified by Victor Manuel Rivas Santos (University of Jaen) 15/07/2004

Constructor Summary

Constructors

Constructor and Description
Rbfn() Creates a instance of rbnf of fixed structure just for test.
Rbfn(double[][] X, int ndatos, int nEnt, int nSal, int nNeuro) Creates a new instance of rbfn from a matrix of instances.
Rbfn(int nNeuro, double[][] X, int ndatos, int nEnt, int nSal) Creates a new instance of rbfn from a matrix of instances.
Rbfn(int nInputs, int nOutputs) Creates a new instance of rbfn

Method Summary

Modifier and Type	Method and Description
void	classificationTest(double[][] X, int ndatos, int[] yieldedResults, int max, int min) Evaluates the net for clasification problem
java.lang.Object	clone() Clones a RBFN neural network
java.lang.String	closestRbf(double[] v) Returns the nearest rbf/neuron to a vector v (pattern)
void	decremental(double[][] X, double[][] Y, int ndatos, double percent, double alfa) Uses a decremental algorithm to buid a net.
double[]	errorRbfn(double[] desiredOutput, double[] netOutput) Computes the difference between the ouput of the net and desired output
double[]	evalRbfn(double[] _input) Passes an input to the net obtaining its output
double	getFitness() Gets the fitness of a RBF.
java.lang.String[]	getIndexes() Returns the list on index of the net neurons.
Rbf	getRbf(java.lang.String id) Gets an RBF from the net given its identifier
void	insertRbf(Rbf rbf) Adds a neuron to the net, assigning an automatic name
void	insertRbf(Rbf rbf, java.lang.String _key) Adds a neuron to the net with a given name
void	LMSTrain(double[][] X, double[][] Y, int ndatos, int iter, double alfa) Uses LMS to train the net.
void	modellingTest(double[][] X, int ndatos, double[] yieldedResults) Evaluates the net for modeling problem
void	modifyRbf(Rbf rbf, java.lang.String idRbf) Changes a neuron in the net
int	numInputs() To get the number of inputs
int	numOutputs() To get the number of outputs
void	paint() Prints net on a stdout
void	paint(java.lang.String _fileName) Prints net on a file.
void	RAN(double[][] X, double[][] Y, int ndatos, double delta, double epsilon, double alfa) Uses RAN algorithm to build a net
int	rbfSize() Returns the number of neurons in the net
void	removeRbf(java.lang.String idRbf) Deletes a neuron from the net
void	setFitness_Cl(double[][] _X, double[][] _Y, int _nDatos, int _nClases) Sets the fitness of a RBFN for classification problems
double	setFitness(double _fitness) Sets the fitness of a RBF.

Methods inherited from class java.lang.Object

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

Constructor Detail

Rbfn

public Rbfn()

Creates a instance of rbnf of fixed structure just for test.

Rbfn

public Rbfn(int nInputs,
            int nOutputs)

Creates a new instance of rbfn

Parameters:

nInputs - Input dimension

nOutputs - Ouput dimension (thus, number of ouput neurons)

Rbfn

public Rbfn(double[][] X,
            int ndatos,
            int nEnt,
            int nSal,
            int nNeuro)

Creates a new instance of rbfn from a matrix of instances. Sets randomly the centres of the neurons and sets its radius taking intro account the maximun distance between centres.

Parameters:

X - Matrix of instances

ndatos - Number of instaces in X

nEnt - Number of imputs of the net

nSal - Number of outputs of the net

nNeuro - Number of hidden neurons the net will have.

Rbfn

public Rbfn(int nNeuro,
            double[][] X,
            int ndatos,
            int nEnt,
            int nSal)

Creates a new instance of rbfn from a matrix of instances. Sets randomly the centres of the neurons and sets its radius taking intro account the average distance between centres.

Parameters:

nNeuro - Number of hidden neurons the net will have.

X - Matrix of instances

ndatos - Number of instaces in X

nEnt - Number of imputs of the net

nSal - Number of outputs of the net

Method Detail

clone

public java.lang.Object clone()

Clones a RBFN neural network

Overrides:

clone in class java.lang.Object

Returns:

a RBFN neural network object.

removeRbf

public void removeRbf(java.lang.String idRbf)

Deletes a neuron from the net

Parameters:

idRbf - Identifier of neuron to delete

insertRbf

public void insertRbf(Rbf rbf)

Adds a neuron to the net, assigning an automatic name

Parameters:

rbf - The neuron to insert

insertRbf

public void insertRbf(Rbf rbf,
                      java.lang.String _key)

Adds a neuron to the net with a given name

Parameters:

rbf - The neuron to insert

_key - Its name

modifyRbf

public void modifyRbf(Rbf rbf,
                      java.lang.String idRbf)

Changes a neuron in the net

Parameters:

idRbf - Identifier of neuron to delete

rbf - Neuron to insert

numInputs

public int numInputs()

To get the number of inputs

Returns:

The number of INPUTS the net has

numOutputs

public int numOutputs()

To get the number of outputs

Returns:

The number of OUTPUTS the net has

rbfSize

public int rbfSize()

Returns the number of neurons in the net

Returns:

A integer that is the number of neurons of the net

getRbf

public Rbf getRbf(java.lang.String id)

Gets an RBF from the net given its identifier

Parameters:

id - RBF's identifier

Returns:

the RBF object with the given identifier.

getIndexes

public java.lang.String[] getIndexes()

Returns the list on index of the net neurons.

Returns:

the list on index of the net neurons.

setFitness_Cl

public void setFitness_Cl(double[][] _X,
                          double[][] _Y,
                          int _nDatos,
                          int _nClases)

Sets the fitness of a RBFN for classification problems

Parameters:

_X - input dataset.

_Y - output dataset.

_nDatos - number of examples (_X length).

_nClases - number of classes (_Y length).

setFitness

public double setFitness(double _fitness)

Sets the fitness of a RBF.

Parameters:

_fitness - The value to be used for the fitness

Returns:

The value _finess itself

getFitness

public double getFitness()

Gets the fitness of a RBF.

Returns:

The value of the finess

evalRbfn

public double[] evalRbfn(double[] _input)

Passes an input to the net obtaining its output

Parameters:

_input - The sample

Returns:

The set of outputs provided by the net's output neurons.

errorRbfn

public double[] errorRbfn(double[] desiredOutput,
                          double[] netOutput)

Computes the difference between the ouput of the net and desired output

Parameters:

desiredOutput - Desired output

netOutput - Outpunt of the net

Returns:

A vector of doubles with the difference between the output of the net and the desired output

closestRbf

public java.lang.String closestRbf(double[] v)

Returns the nearest rbf/neuron to a vector v (pattern)

Parameters:

v - vector

Returns:

The index (or key) of the closest neuron

RAN

public void RAN(double[][] X,
                double[][] Y,
                int ndatos,
                double delta,
                double epsilon,
                double alfa)

Uses RAN algorithm to build a net

Parameters:

X - matrix of inputs instances

Y - matrix of outputs instances

ndatos - Number of instances

epsilon - minimun error to introduce a new RBF

delta - minimun distance to introduce a new RBF

alfa - learning factor when a new unit is not allocated

decremental

public void decremental(double[][] X,
                        double[][] Y,
                        int ndatos,
                        double percent,
                        double alfa)

Uses a decremental algorithm to buid a net. After initializing and training (with LMS) a net with several neurons, the algorithm in the major loop deletes the neurons with the lowest weight and train the net.

Parameters:

X - matrix of inputs instances

Y - matrix of outputs instances

ndatos - Number of instances

percent - Percent under the average of the weights to delete a neuron

alfa - Learnig factor of LMS algorithm

LMSTrain

public void LMSTrain(double[][] X,
                     double[][] Y,
                     int ndatos,
                     int iter,
                     double alfa)

Uses LMS to train the net.

Parameters:

X - matrix of inputs instances

Y - matrix of outputs instances

ndatos - Number of instances

iter - Number of times the set of samples will be used.

alfa - Learning factor.

modellingTest

public void modellingTest(double[][] X,
                          int ndatos,
                          double[] yieldedResults)

Evaluates the net for modeling problem

Parameters:

X - matrix of inputs instances

ndatos - Number of instances

yieldedResults - Vector of results of the evaluation

classificationTest

public void classificationTest(double[][] X,
                               int ndatos,
                               int[] yieldedResults,
                               int max,
                               int min)

Evaluates the net for clasification problem

Parameters:

X - matrix of inputs instances

ndatos - Number of instances

yieldedResults - Vector of results of the evaluation

max - Class maximun identifier

min - Class minimun identifier

paint

public void paint()

Prints net on a stdout

paint

public void paint(java.lang.String _fileName)

Prints net on a file.

Parameters:

_fileName - Name of the file.