edu.iu.dsc.tws.examples.ml.svm.sgd.pegasos

Class PegasosSgdSvm

java.lang.Object

edu.iu.dsc.tws.examples.ml.svm.sgd.SgdSvm

edu.iu.dsc.tws.examples.ml.svm.sgd.pegasos.PegasosSgdSvm

All Implemented Interfaces:

java.io.Serializable

public class PegasosSgdSvm
extends SgdSvm
implements java.io.Serializable

This is the Pegasos based Sgd SVM Class Here we have only implemented the linear kernel based algorithm TODO : Dynamic Model based update (use Model extended classes to initialize this class) TODO : Implement Gaussian Kernel based SGD SVM

See Also:

Serialized Form

Field Summary

Fields inherited from class edu.iu.dsc.tws.examples.ml.svm.sgd.SgdSvm

alpha, dataLoadingTime, isInvalid, iterations, samples, testingTime, trainingTime, w, x, xBatch, y, yBatch

Constructor Summary

Constructors

Constructor and Description
PegasosSgdSvm(double[] w, double[][] x, double[] y, double alpha, int iterations, int features)
PegasosSgdSvm(double[] w, double[] x, double y, double alpha, int iterations)
PegasosSgdSvm(double[] x, double y, double alpha, int iterations)
PegasosSgdSvm(double[] w, double alpha, int iterations, int features)

Method Summary

Modifier and Type	Method and Description
double[]	getW()
void	iterativeSgd(double[] w, double[][] x, double[] y) This is the iterative Sgd based SVM for Linear Kernel
void	iterativeTaskSgd(double[] w1, double[][] x1, double[] y1)
<T> void	onlineDynamicSGD(T[] w1, T[] x1, T y1)
void	onlineSGD(double[] w, double[] x, double y) This is the Online Sgd based SVM for Linear Kernel
void	setW(double[] w)
void	sgd() Deprecated. Use iterativeSGD for batch mode training Use onlineSGD for streaming mode training

Methods inherited from class java.lang.Object

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

Constructor Detail

PegasosSgdSvm

public PegasosSgdSvm(double[] x,
                     double y,
                     double alpha,
                     int iterations)

PegasosSgdSvm

public PegasosSgdSvm(double[] w,
                     double[] x,
                     double y,
                     double alpha,
                     int iterations)

PegasosSgdSvm

public PegasosSgdSvm(double[] w,
                     double alpha,
                     int iterations,
                     int features)

PegasosSgdSvm

public PegasosSgdSvm(double[] w,
                     double[][] x,
                     double[] y,
                     double alpha,
                     int iterations,
                     int features)

Method Detail

sgd

@Deprecated
public void sgd()
                     throws NullDataSetException,
                            MatrixMultiplicationException

Deprecated. Use iterativeSGD for batch mode training Use onlineSGD for streaming mode training

This method is deprected

Specified by:

sgd in class SgdSvm

Throws:

NullDataSetException

MatrixMultiplicationException

iterativeSgd

public void iterativeSgd(double[] w,
                         double[][] x,
                         double[] y)
                  throws NullDataSetException,
                         MatrixMultiplicationException

This is the iterative Sgd based SVM for Linear Kernel

Specified by:

iterativeSgd in class SgdSvm

Parameters:

w - initial weight vector with x.length dimension

x - all data points x.length = samples x[0].length == w.length must be true

y - all labels per each data point

Throws:

NullDataSetException

MatrixMultiplicationException

iterativeTaskSgd

public void iterativeTaskSgd(double[] w1,
                             double[][] x1,
                             double[] y1)
                      throws NullDataSetException,
                             MatrixMultiplicationException

Specified by:

iterativeTaskSgd in class SgdSvm

Throws:

NullDataSetException

MatrixMultiplicationException

onlineSGD

public void onlineSGD(double[] w,
                      double[] x,
                      double y)
               throws NullDataSetException,
                      MatrixMultiplicationException

This is the Online Sgd based SVM for Linear Kernel

Specified by:

onlineSGD in class SgdSvm

Parameters:

w - initial weight vector with x.length dimension

x - single data point x.length = 1 x.length == w.length must be true

y - label of the data point

Throws:

NullDataSetException

MatrixMultiplicationException

onlineDynamicSGD

public <T> void onlineDynamicSGD(T[] w1,
                                 T[] x1,
                                 T y1)

Specified by:

onlineDynamicSGD in class SgdSvm

getW

public double[] getW()

Overrides:

getW in class SgdSvm

setW

public void setW(double[] w)

Overrides:

setW in class SgdSvm