SciLib.Neuro.SVM
Class BSVM

java.lang.Object
  SciLib.Neuro.SVM.SVM
      SciLib.Neuro.SVM.BSVM

All Implemented Interfaces:

SMO

public class BSVM
extends SVM
implements SMO

Field Summary

protected int	c1
protected int	c2

Fields inherited from class SciLib.Neuro.SVM.SVM

alpha, B, b_low, b_up, binary, C, eps, error_cache, file, GACV1, i_low, i_up, JHB, K, M, N, patterns, sparse, sv, svIndex, targets, tolerance

Constructor Summary

BSVM(FileHandler file, boolean transf, int c1)
Constructor

BSVM(FileHandler file, Kernel K, boolean transf, int c1)
Constructor

BSVM(matrix patterns, integerVector targets, Kernel K)
Constructor

BSVM(matrix patterns, integerVector targets, Kernel K, boolean transf, int c1)
Constructor

BSVM(matrix patterns, integerVector targets, Kernel K, int c1, int c2)
Constructor

Method Summary

void	crossValidation(int k, double startLog2C, int cs, double startLambda, int ls, double a, double b) cross-validation
void	doneTransf()
boolean	doneTransform()
int	getC1()
int	getC2()
void	LOOValidation(int type, double[] cPar, double[] lambdaPar, double a, double b)
void	LOOValidation(int type, double startLog2C, int cs, double startLambda, int ls, double a, double b) Leave-one-out-validation
int	misclassified(matrix p, integerVector t)
double	testing(FileHandler testFile) test svm without scaling method
double	testing(FileHandler testFile, double a, double b) test svm with scaling method
double	testing(FileHandler testFile, java.lang.String alphaFile) test svm without scaling method
double	testing(FileHandler testFile, java.lang.String alphaFile, double a, double b) test svm with scaling method
void	training(java.lang.String alphaFile) the training algorithm with no cross validation and no scaling
void	training(java.lang.String alphaFile, double a, double b) training svm with the data scaled between a, b ( a < b )

Methods inherited from class SciLib.Neuro.SVM.SVM

computeB, f, f, getAlpha, getB, getC, getEpsi, getFile, getKernel, getNoOfSV, getPatterns, getSV, getSVIndex, getTargets, getTolerance, initialize, initiateParameter, isBinary, isInI0I1I2, isInI0I3I4, isInI1I2, isInI3I4, isSparse, makeTargets, readSVs, setB, setC, setEpsi, setFile, setKernel, setN, setTolerance, writeReport, writeResultToFile

Methods inherited from class java.lang.Object

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

Field Detail

c1

protected int c1

c2

protected int c2

Constructor Detail

BSVM

public BSVM(FileHandler file,
            Kernel K,
            boolean transf,
            int c1)

Constructor

Parameters:

file - the file that handles training data, FileHandler

K - the kernel of SVM, Kernel

transf - convert the target data to the form of 1 and -1, boolean

c1 - will be convert to 1, int

BSVM

public BSVM(FileHandler file,
            boolean transf,
            int c1)

Constructor

Parameters:

file - the file that handles training data, FileHandler

transf - convert the target data to the form of 1 and -1, boolean

c1 - will be convert to 1, int

BSVM

public BSVM(matrix patterns,
            integerVector targets,
            Kernel K,
            boolean transf,
            int c1)

Constructor

Parameters:

patterns - the training data, matrix

targets - labels of the training data, integerVector

K - the kernel of SVM, Kernel

transf - convert the target data to the form of 1 and -1, boolean

c1 - will be convert to 1, int

BSVM

public BSVM(matrix patterns,
            integerVector targets,
            Kernel K,
            int c1,
            int c2)

Constructor

Parameters:

patterns - the training data, matrix

targets - labels of the training data, integerVector

K - the kernel of SVM, Kernel

c1 - will be convert to 1, int

c2 - will be convert to -1, int

BSVM

public BSVM(matrix patterns,
            integerVector targets,
            Kernel K)

Constructor

Parameters:

patterns - the training data, matrix

targets - labels of the training data, integerVector

K - the kernel of SVM, Kernel

Method Detail

doneTransform

public boolean doneTransform()

doneTransf

public void doneTransf()

getC1

public int getC1()

getC2

public int getC2()

training

public void training(java.lang.String alphaFile)

the training algorithm with no cross validation and no scaling

Specified by:

training in interface SMO

Overrides:

training in class SVM

Parameters:

alphaFile - fileName for store the support vectors

training

public void training(java.lang.String alphaFile,
                     double a,
                     double b)

training svm with the data scaled between a, b ( a < b )

Specified by:

training in interface SMO

Overrides:

training in class SVM

Parameters:

alphaFile - fileName for store the support vectors

a - the lower bound of scaled data

b - the upper bound of scaled data

crossValidation

public void crossValidation(int k,
                            double startLog2C,
                            int cs,
                            double startLambda,
                            int ls,
                            double a,
                            double b)

cross-validation

Specified by:

crossValidation in interface SMO

Overrides:

crossValidation in class SVM

Parameters:

k - k-fold cross validation k > 1

startLog2C - log2 of the start value of C, double

cs - the size of the set that contains all C, int

startLambda - log2 of the start value of lambda, double

ls - the size of the set that contains all lambdas, int

a - the lower bound of scaled data

b - the upper bound of scaled data

misclassified

public int misclassified(matrix p,
                         integerVector t)

Specified by:

misclassified in interface SMO

Overrides:

misclassified in class SVM

testing

public double testing(FileHandler testFile)

test svm without scaling method

Specified by:

testing in interface SMO

Overrides:

testing in class SVM

Parameters:

testFile - FileHandler handles the test file data

testing

public double testing(FileHandler testFile,
                      double a,
                      double b)

test svm with scaling method

Specified by:

testing in interface SMO

Overrides:

testing in class SVM

Parameters:

testFile - FileHandler handles the test file data

a - the lower bound of scaled data

b - the upper bound of scaled data

testing

public double testing(FileHandler testFile,
                      java.lang.String alphaFile)

test svm without scaling method

Specified by:

testing in interface SMO

Overrides:

testing in class SVM

Parameters:

testFile - FileHandler handles the test file data

alphaFile - the file name of the file that stores support vectors

testing

public double testing(FileHandler testFile,
                      java.lang.String alphaFile,
                      double a,
                      double b)

test svm with scaling method

Specified by:

testing in interface SMO

Overrides:

testing in class SVM

Parameters:

testFile - FileHandler handles the test file data

alphaFile - the file name of the file that stores support vectors

a - the lower bound of scaled data

b - the upper bound of scaled data

LOOValidation

public void LOOValidation(int type,
                          double startLog2C,
                          int cs,
                          double startLambda,
                          int ls,
                          double a,
                          double b)

Leave-one-out-validation

Specified by:

LOOValidation in interface SMO

Overrides:

LOOValidation in class SVM

Parameters:

type - method of LOO validation, constant JHB, GACV1. etc.

startLog2C - log2 of the start value of C, double

cs - the size of the set that contains all C, int

startLambda - log2 of the start value of lambda, double

ls - the size of the set that contains all lambdas, int

a - the lower bound of scaled data

b - the upper bound of scaled data

LOOValidation

public void LOOValidation(int type,
                          double[] cPar,
                          double[] lambdaPar,
                          double a,
                          double b)

Specified by:

LOOValidation in interface SMO

Overrides:

LOOValidation in class SVM