keel.Algorithms.Neural_Networks.RBFN_CL

Class Rbfn

java.lang.Object

keel.Algorithms.Neural_Networks.RBFN_CL.Rbfn

public class Rbfn
extends java.lang.Object

This class codified a Radial Basis Function Network

Since:

JDK1.5

Version:

1.0

Author:

Writen by Antonio J. Rivera Rivas (University of Jaen) 03/03/2004, Modified by Victor Rivas (University of Jaen) 23/06/2006, Modified by Maria Dolores Perez Godoy (University of Jaen) 17/12/2008

Constructor Summary

Constructors

Constructor and Description
Rbfn() Creates a instance of rbnf of fixed structure just for test.
Rbfn(double[][] X, int ndata, int nInpt, int nOutp, int nNeuro) Creates a new instance of rbfn from a matrix of instances.
Rbfn(int nNeuro, double[][] X, int ndata, int nInpt, int nOutpl) Creates a new instance of rbfn from a matrix of instances.
Rbfn(int numInpt, int numOutp) Creates a new instance of rbfn

Method Summary

Modifier and Type	Method and Description
void	decremental(double[][] X, double[][] Y, int ndata, double percent, double alfa) Uses a decremental algorithm to buid a net.
double[]	errorRbfn(double[] realOutput, double[] netOutput) Computes the difference between the ouput of the net and desired outpunt
protected double	euclideanDist(double[] v1, double[] v2) Computes the euclidean distance between 2 vectors
double[]	evaluationRbfn(double[] _input) Passes an input to the net obtaining its output
java.lang.String[]	getIndex() 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
void	modifyRbf(Rbf rbf, java.lang.String idRbf) Changes a neuron in the net
void	printRbfn() Prints net on a stdout
void	printRbfn(java.lang.String _fileName) Prints net on a file.
void	RAN(double[][] X, double[][] Y, int ndata, double delta, double epsilon, double alfa) Uses RAN algorithm to build a net
java.lang.String	rbfNearest(double[] v) Returns the nearest rbf/neuron to a vector v (patron)
void	removeRbf(java.lang.String idRbf) Deletes a neuron of the net
int	size() Returns the number of neurons in the net
void	testClasification(double[][] X, int ndata, int[] obtained, int max, int min) Evaluates the net for clasification problem
void	testModeling(double[][] X, int ndata, double[] obtained) Evaluates the net for modeling problem
void	trainLMS(double[][] X, double[][] Y, int ndata, int iter, double alfa) Uses LMS to train the net.

Methods inherited from class java.lang.Object

clone, 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 numInpt,
            int numOutp)

Creates a new instance of rbfn

Parameters:

numInpt - Input dimension

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

Rbfn

public Rbfn(double[][] X,
            int ndata,
            int nInpt,
            int nOutp,
            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

ndata - Number of instaces in X

nInpt - Number of imputs of the net

nOutp - Number of outputs of the net

nNeuro - Number of hidden neurons the net will have.

Rbfn

public Rbfn(int nNeuro,
            double[][] X,
            int ndata,
            int nInpt,
            int nOutpl)

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

ndata - Number of instaces in X

nInpt - Number of imputs of the net

nOutpl - Number of outputs of the net

Method Detail

euclideanDist

protected double euclideanDist(double[] v1,
                               double[] v2)

Computes the euclidean distance between 2 vectors

Parameters:

v1 - A vector

v2 - A vector

Returns:

a double with the euclidean distance between v1 and v2

removeRbf

public void removeRbf(java.lang.String idRbf)

Deletes a neuron of the net

Parameters:

idRbf - Identifier of neuron to delete

insertRbf

public void insertRbf(Rbf rbf)

Adds a neuron to the net

Parameters:

rbf - The neuron to insert

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

size

public int size()

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 whose identified is id

getIndex

public java.lang.String[] getIndex()

Returns the list on index of the net neurons.

Returns:

a string with the list on index of the net neurons

evaluationRbfn

public double[] evaluationRbfn(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[] realOutput,
                          double[] netOutput)

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

Parameters:

realOutput - 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

rbfNearest

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

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

Parameters:

v - vector

Returns:

the identifier of the nearest RBF to v

RAN

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

Uses RAN algorithm to build a net

Parameters:

X - matrix of inputs instances

Y - matrix of outputs instances

ndata - 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 ndata,
                        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

ndata - Number of instances

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

alfa - Learnig factor of LMS algorithm

trainLMS

public void trainLMS(double[][] X,
                     double[][] Y,
                     int ndata,
                     int iter,
                     double alfa)

Uses LMS to train the net.

Parameters:

X - matrix of inputs instances

Y - matrix of outputs instances

ndata - Number of instances

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

alfa - Learning factor.

testModeling

public void testModeling(double[][] X,
                         int ndata,
                         double[] obtained)

Evaluates the net for modeling problem

Parameters:

X - matrix of inputs instances

ndata - Number of instances

obtained - Vector of results of the evaluation

testClasification

public void testClasification(double[][] X,
                              int ndata,
                              int[] obtained,
                              int max,
                              int min)

Evaluates the net for clasification problem

Parameters:

X - matrix of inputs instances

ndata - Number of instances

obtained - Vector of results of the evaluation

max - Class maximun identifier

min - Class minimun identifier

printRbfn

public void printRbfn()

Prints net on a stdout

printRbfn

public void printRbfn(java.lang.String _fileName)

Prints net on a file.

Parameters:

_fileName - Name of the file.