SciLib.Neuro
Class RbfNet

java.lang.Object
  SciLib.Neuro.Nnet
      SciLib.Neuro.RbfNet

public class RbfNet
extends Nnet

This class defines A Radial Basis Function Network.

Field Summary

integerVector[]	c contains K-cluster indexes
protected matrix	G
protected matrix	G0
static int	GENERALIZATIONRBFNW
static int	INTERPOLATIONRBFNW
protected double	lambda
static int	RADIALBFNW
protected double	radius
matrix	t The set of centers
protected int	type

Fields inherited from class SciLib.Neuro.Nnet

averageError, DesiredOutputMatrix, errorThreshold, g, hidden, hiddenSize, inSize, learningRate, momentum, numberOfHiddenLayers, NumberOfTestPatterns, NumberOfTrainingPatterns, output, outSize, testingData, testingDesiredOutput, trainingData, trainingDesiredOutput

Constructor Summary

RbfNet()
Make a default Radial basis function network

RbfNet(int is, int os, int hs, int type)
Make a Radial basis function network with specific size and type

RbfNet(int is, int os, int hs, int type, Activation phi)
Make a Radial basis function network with specific size, type and activation function

Method Summary

vector	approximate(Function f, vector x)
double	classify(int type, integerVector result) classify the data from a test file
void	computeOutput(vector x) Compute the output from an input vector x
double	computePhiAt(vector x, int i) Compute phi(x_i)
double	computeRadius() Compute the radius = M/d^2
void	generateCenters(int method, int[] centers) generate centers for a radial basis function network
void	generateG0() Generate the matrix G0
void	generateGreenMatrix() Generate the Green matrix G
void	generateKCenters(int[] centers) Generate k fixed centers
void	generateRandomlyCenters() Genrate k centers randomly
matrix	getGreenMatrix() get Green matrix G
java.lang.String	getType() Get the radial basis function network type
void	initialize(double radius, double lambda, int type) initialize a radial basis function network
void	initialize(double radius, double lambda, int type, Activation phi) initialize a radial basis function network
void	interpolation(double lambda) Interpolation method
vector	makeTrainingDesiredOutputVector(int i) make a current training desired output vector
void	training(int trainingType, int[] c) Training the network
void	training(int trainingType, int[] c, double lambda) Training the network with a generalization parameter lambda

Methods inherited from class SciLib.Neuro.Nnet

generate, getErrorThreshold, getHiddenLayers, gethiddenSize, getInputSize, getLearningRate, getMomentumPar, getnumberOfHiddenLayers, getNumberOfTestPatterns, getNumberOfTrainingPatterns, getOutputLayer, getOutputSize, getTrainingData, getTrainingDesiredOutput, makeDesiredOutputMatrix, makeDesiredOutputMatrix, makeTrainingSet, readTestingData, readTrainingData, readWeight, setActivation, setErrorThreshold, setInputSize, setLearningRate, setMomentumPar, setNumberOfTestPatterns, setNumberOfTrainingPatterns, setOutputSize, setSize, writeWeight

Methods inherited from class java.lang.Object

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

Field Detail

RADIALBFNW

public static final int RADIALBFNW

See Also:

Constant Field Values

INTERPOLATIONRBFNW

public static final int INTERPOLATIONRBFNW

See Also:

Constant Field Values

GENERALIZATIONRBFNW

public static final int GENERALIZATIONRBFNW

See Also:

Constant Field Values

G

protected matrix G

G0

protected matrix G0

lambda

protected double lambda

radius

protected double radius

type

protected int type

t

public matrix t

The set of centers

c

public integerVector[] c

contains K-cluster indexes

Constructor Detail

RbfNet

public RbfNet()

Make a default Radial basis function network

RbfNet

public RbfNet(int is,
              int os,
              int hs,
              int type)

Make a Radial basis function network with specific size and type

Parameters:

is - An integer value, the input size

os - An integer value, the output size

hs - An integer value, the size of the hidden layer

type - An integer value, the type of radial basis function network

RbfNet

public RbfNet(int is,
              int os,
              int hs,
              int type,
              Activation phi)

Make a Radial basis function network with specific size, type and activation function

Parameters:

is - An integer value, the input size

os - An integer value, the output size

hs - An integer value, the size of the hidden layer

type - An integer value, the type of radial basis function network

phi - An Activation object

Method Detail

getGreenMatrix

public matrix getGreenMatrix()

get Green matrix G

Returns:

a matrix

getType

public java.lang.String getType()

Get the radial basis function network type

Returns:

A String

initialize

public void initialize(double radius,
                       double lambda,
                       int type)

initialize a radial basis function network

Parameters:

radius - A double value

lambda - A double value

type - A double value

initialize

public void initialize(double radius,
                       double lambda,
                       int type,
                       Activation phi)

initialize a radial basis function network

Parameters:

radius - A double value

lambda - A double value

type - A double value

phi - An Activation function

computePhiAt

public double computePhiAt(vector x,
                           int i)

Compute phi(x_i)

Parameters:

x - A vector

i - An integer value

computeRadius

public double computeRadius()

Compute the radius = M/d^2

generateCenters

public void generateCenters(int method,
                            int[] centers)

generate centers for a radial basis function network

Parameters:

method - 0 for randomly, 1 for k-mean clustering and 0 for fixed centers

centers - An integer array of indexes of fixed centers. It will be null for other methods

generateKCenters

public void generateKCenters(int[] centers)

Generate k fixed centers

Parameters:

centers - An integer array of indexes of fixed centers.

generateRandomlyCenters

public void generateRandomlyCenters()

Genrate k centers randomly

generateGreenMatrix

public void generateGreenMatrix()

Generate the Green matrix G

generateG0

public void generateG0()

Generate the matrix G0

computeOutput

public void computeOutput(vector x)

Compute the output from an input vector x

Parameters:

x - A vector

makeTrainingDesiredOutputVector

public vector makeTrainingDesiredOutputVector(int i)

make a current training desired output vector

training

public void training(int trainingType,
                     int[] c)

Training the network

Parameters:

trainingType - the training method, 0 for randomly, 1 for k-mean clustering and 0 for fixed centers

c - An integer array of indexes of fixed centers. It will be null for other methods

interpolation

public void interpolation(double lambda)

Interpolation method

training

public void training(int trainingType,
                     int[] c,
                     double lambda)

Training the network with a generalization parameter lambda

Parameters:

trainingType - the training method, 0 for randomly, 1 for k-mean clustering and 0 for fixed centers

c - An integer array of indexes of fixed centers. It will be null for other methods

lambda - A double value

classify

public double classify(int type,
                       integerVector result)

classify the data from a test file

Parameters:

type - An integer value, the classification method

result - An integerVector contains the classification result

Returns:

a double value, the correct classification in percent

approximate

public vector approximate(Function f,
                          vector x)