RegSMOImproved

java.lang.Object
- keel.Algorithms.SVM.SMO.supportVector.RegOptimizer
- - keel.Algorithms.SVM.SMO.supportVector.RegSMO
  - - keel.Algorithms.SVM.SMO.supportVector.RegSMOImproved

All Implemented Interfaces:: java.io.Serializable, OptionHandler, TechnicalInformationHandler

public class RegSMOImproved
extends RegSMO
implements TechnicalInformationHandler

Learn SVM for regression using SMO with Shevade, Keerthi, et al. adaption of the stopping criterion.

For more information see:

S.K. Shevade, S.S. Keerthi, C. Bhattacharyya, K.R.K. Murthy: Improvements to the SMO Algorithm for SVM Regression. In: IEEE Transactions on Neural Networks, 1999.

S.K. Shevade, S.S. Keerthi, C. Bhattacharyya, K.R.K. Murthy (1999). Improvements to the SMO Algorithm for SVM Regression. Control Division, Dept. of Mechanical Engineering.

BibTeX:

 @inproceedings{Shevade1999,
    author = {S.K. Shevade and S.S. Keerthi and C. Bhattacharyya and K.R.K. Murthy},
    booktitle = {IEEE Transactions on Neural Networks},
    title = {Improvements to the SMO Algorithm for SVM Regression},
    year = {1999},
    PS = {http://guppy.mpe.nus.edu.sg/~mpessk/svm/ieee_smo_reg.ps.gz}
 }
 
 @techreport{Shevade1999,
    address = {Control Division, Dept. of Mechanical Engineering},
    author = {S.K. Shevade and S.S. Keerthi and C. Bhattacharyya and K.R.K. Murthy},
    institution = {National University of Singapore},
    number = {CD-99-16},
    title = {Improvements to the SMO Algorithm for SVM Regression},
    year = {1999},
    PS = {http://guppy.mpe.nus.edu.sg/~mpessk/svm/smoreg_mod.ps.gz}
 }

Valid options are:

 -T <double>
  The tolerance parameter for checking the stopping criterion.
  (default 0.001)

 -V
  Use variant 1 of the algorithm when true, otherwise use variant 2.
  (default true)

 -P <double>
  The epsilon for round-off error.
  (default 1.0e-12)

 -L <double>
  The epsilon parameter in epsilon-insensitive loss function.
  (default 1.0e-3)

 -W <double>
  The random number seed.
  (default 1)

Version:: $Revision: 1.1 $
Author:: Remco Bouckaert (remco@cs.waikato.ac.nz,rrb@xm.co.nz)
See Also:: Serialized Form

Field Summary

Fields
Modifier and Type	Field and Description
`static int`	`I0` Configuration flags
`static int`	`I0a` Configuration flags
`static int`	`I0b` Configuration flags
`static int`	`I1` Configuration flags
`static int`	`I2` Configuration flags
`static int`	`I3` Configuration flags
`protected double`	`m_bLow` b.up and b.low boundaries used to determine stopping criterion
`protected double`	`m_bUp` b.up and b.low boundaries used to determine stopping criterion
`protected SMOset`	`m_I0` The different sets used by the algorithm.
`protected int`	`m_iLow` index of the instance that gave us b.up and b.low
`protected int[]`	`m_iSet` Index set {i: 0 < m_alpha[i] < C \|\| 0 < m_alphaStar[i] < C}}
`protected int`	`m_iUp` index of the instance that gave us b.up and b.low

Fields inherited from class keel.Algorithms.SVM.SMO.supportVector.RegSMO
m_alpha1, m_alpha1Star, m_alpha2, m_alpha2Star, m_Del, m_eps

Fields inherited from class keel.Algorithms.SVM.SMO.supportVector.RegOptimizer
m_alpha, m_alphaStar, m_b, m_bModelBuilt, m_C, m_classIndex, m_data, m_epsilon, m_kernel, m_nCacheHits, m_nEvals, m_nInstances, m_nSeed, m_random, m_sparseIndices, m_sparseWeights, m_supportVectors, m_SVM, m_target, m_weights

Constructor Summary

Constructors
Constructor and Description

RegSMOImproved()

Constructors
Constructor and Description
`RegSMOImproved()`

Method Summary

All Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`void`	`buildClassifier(Instances instances)` learn SVM parameters from data using Keerthi's SMO algorithm.
`protected int`	`examineExample(int i2)` parameters correspond to pseudocode from paper.
`java.lang.String[]`	`getOptions()` Gets the current settings of the object.
`TechnicalInformation`	`getTechnicalInformation()` Returns an instance of a TechnicalInformation object, containing detailed information about the technical background of this class, e.g., paper reference or book this class is based on.
`double`	`getTolerance()` returns the current tolerance
`java.lang.String`	`globalInfo()` Returns a string describing the object
`protected void`	`init(Instances data)` initialize various variables before starting the actual optimizer
`boolean`	`isUseVariant1()` Whether variant 1 is used
`java.util.Enumeration`	`listOptions()` Returns an enumeration describing the available options
`protected void`	`optimize1()` use variant 1 of Shevade's et al.s paper
`protected void`	`optimize2()` use variant 2 of Shevade's et al.s paper
`void`	`setOptions(java.lang.String[] options)` Parses a given list of options.
`void`	`setTolerance(double d)` sets the tolerance
`void`	`setUseVariant1(boolean b)` Sets whether to use variant 1
`protected int`	`takeStep(int i1, int i2, double alpha2, double alpha2Star, double phi2)` takeStep method from Shevade et al.s paper.
`java.lang.String`	`toleranceTipText()` Returns the tip text for this property
`protected void`	`updateBoundaries(int i2, double F2)` updates boundaries bLow and bHi and corresponding indexes
`protected void`	`updateIndexSetFor(int i, double C)` updates the index sets I0a, IOb, I1, I2 and I3 for vector i
`java.lang.String`	`useVariant1TipText()` Returns the tip text for this property
`protected void`	`wrapUp()` wrap up various variables to save memeory and do some housekeeping after optimization has finished.

Methods inherited from class keel.Algorithms.SVM.SMO.supportVector.RegSMO
epsilonTipText, findOptimalPointOnLine, getEpsilon, optimize, secondChoiceHeuristic, setEpsilon

Methods inherited from class keel.Algorithms.SVM.SMO.supportVector.RegOptimizer
epsilonParameterTipText, getCacheHits, getEpsilonParameter, getKernelEvaluations, getScore, getSeed, modelBuilt, seedTipText, setEpsilonParameter, setSeed, setSVMReg, SVMOutput, SVMOutput

Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

- Field Detail
  - I0
```
public static final int I0
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - I0a
```
public static final int I0a
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - I0b
```
public static final int I0b
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - I1
```
public static final int I1
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - I2
```
public static final int I2
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - I3
```
public static final int I3
```
    Configuration flags
    
    See Also:
    
    Constant Field Values
  - m_I0
```
protected SMOset m_I0
```
    The different sets used by the algorithm.
  - m_iSet
```
protected int[] m_iSet
```
    Index set {i: 0 < m_alpha[i] < C || 0 < m_alphaStar[i] < C}}
  - m_bUp
```
protected double m_bUp
```
    b.up and b.low boundaries used to determine stopping criterion
  - m_bLow
```
protected double m_bLow
```
    b.up and b.low boundaries used to determine stopping criterion
  - m_iUp
```
protected int m_iUp
```
    index of the instance that gave us b.up and b.low
  - m_iLow
```
protected int m_iLow
```
    index of the instance that gave us b.up and b.low
- Constructor Detail
  - RegSMOImproved
```
public RegSMOImproved()
```
- Method Detail
  - globalInfo
```
public java.lang.String globalInfo()
```
    Returns a string describing the object
    
    Overrides:
    
    globalInfo in class RegSMO
    
    Returns:
    
    a description suitable for displaying in the explorer/experimenter gui
  - getTechnicalInformation
```
public TechnicalInformation getTechnicalInformation()
```
    Returns an instance of a TechnicalInformation object, containing detailed information about the technical background of this class, e.g., paper reference or book this class is based on.
    
    Specified by:
    
    getTechnicalInformation in interface TechnicalInformationHandler
    
    Overrides:
    
    getTechnicalInformation in class RegSMO
    
    Returns:
    
    the technical information about this class
  - listOptions
```
public java.util.Enumeration listOptions()
```
    Returns an enumeration describing the available options
    
    Specified by:
    
    listOptions in interface OptionHandler
    
    Overrides:
    
    listOptions in class RegSMO
    
    Returns:
    
    an enumeration of all the available options
  - setOptions
```
public void setOptions(java.lang.String[] options)
                throws java.lang.Exception
```
    Parses a given list of options.
    Valid options are:
```
 -T <double>
  The tolerance parameter for checking the stopping criterion.
  (default 0.001)
```
```
 -V
  Use variant 1 of the algorithm when true, otherwise use variant 2.
  (default true)
```
```
 -P <double>
  The epsilon for round-off error.
  (default 1.0e-12)
```
```
 -L <double>
  The epsilon parameter in epsilon-insensitive loss function.
  (default 1.0e-3)
```
```
 -W <double>
  The random number seed.
  (default 1)
```
    Specified by:
    
    setOptions in interface OptionHandler
    
    Overrides:
    
    setOptions in class RegSMO
    
    Parameters:
    
    options - the list of options as an array of strings
    
    Throws:
    
    java.lang.Exception - if an option is not supported
  - getOptions
```
public java.lang.String[] getOptions()
```
    Gets the current settings of the object.
    
    Specified by:
    
    getOptions in interface OptionHandler
    
    Overrides:
    
    getOptions in class RegSMO
    
    Returns:
    
    an array of strings suitable for passing to setOptions
  - toleranceTipText
```
public java.lang.String toleranceTipText()
```
    Returns the tip text for this property
    
    Returns:
    
    a description suitable for displaying in the explorer/experimenter gui
  - getTolerance
```
public double getTolerance()
```
    returns the current tolerance
    
    Returns:
    
    the tolerance
  - setTolerance
```
public void setTolerance(double d)
```
    sets the tolerance
    
    Parameters:
    
    d - the new tolerance
  - useVariant1TipText
```
public java.lang.String useVariant1TipText()
```
    Returns the tip text for this property
    
    Returns:
    
    a description suitable for displaying in the explorer/experimenter gui
  - isUseVariant1
```
public boolean isUseVariant1()
```
    Whether variant 1 is used
    
    Returns:
    
    true if variant 1 is used
  - setUseVariant1
```
public void setUseVariant1(boolean b)
```
    Sets whether to use variant 1
    
    Parameters:
    
    b - if true then variant 1 is used
  - takeStep
```
protected int takeStep(int i1,
                       int i2,
                       double alpha2,
                       double alpha2Star,
                       double phi2)
                throws java.lang.Exception
```
    takeStep method from Shevade et al.s paper. parameters correspond to pseudocode from paper.
    
    Overrides:
    
    takeStep in class RegSMO
    
    Parameters:
    
    i1 -
    
    i2 -
    
    alpha2 -
    
    alpha2Star -
    
    phi2 -
    
    Returns:
    
    Throws:
    
    java.lang.Exception
  - updateIndexSetFor
```
protected void updateIndexSetFor(int i,
                                 double C)
                          throws java.lang.Exception
```
    updates the index sets I0a, IOb, I1, I2 and I3 for vector i
    
    Parameters:
    
    i - index of vector
    
    C - capacity for vector i
    
    Throws:
    
    java.lang.Exception
  - updateBoundaries
```
protected void updateBoundaries(int i2,
                                double F2)
```
    updates boundaries bLow and bHi and corresponding indexes
    
    Parameters:
    
    i2 - index of vector
    
    F2 - error of vector i2
  - examineExample
```
protected int examineExample(int i2)
                      throws java.lang.Exception
```
    parameters correspond to pseudocode from paper.
    
    Overrides:
    
    examineExample in class RegSMO
    
    Parameters:
    
    i2 - index of candidate
    
    Returns:
    
    Throws:
    
    java.lang.Exception
  - init
```
protected void init(Instances data)
             throws java.lang.Exception
```
    initialize various variables before starting the actual optimizer
    
    Overrides:
    
    init in class RegSMO
    
    Parameters:
    
    data - data set used for learning
    
    Throws:
    
    java.lang.Exception - if something goes wrong
  - optimize1
```
protected void optimize1()
                  throws java.lang.Exception
```
    use variant 1 of Shevade's et al.s paper
    
    Throws:
    
    java.lang.Exception - if something goes wrong
  - optimize2
```
protected void optimize2()
                  throws java.lang.Exception
```
    use variant 2 of Shevade's et al.s paper
    
    Throws:
    
    java.lang.Exception - if something goes wrong
  - wrapUp
```
protected void wrapUp()
               throws java.lang.Exception
```
    wrap up various variables to save memeory and do some housekeeping after optimization has finished.
    
    Overrides:
    
    wrapUp in class RegSMO
    
    Throws:
    
    java.lang.Exception - if something goes wrong
  - buildClassifier
```
public void buildClassifier(Instances instances)
                     throws java.lang.Exception
```
    learn SVM parameters from data using Keerthi's SMO algorithm. Subclasses should implement something more interesting.
    
    Overrides:
    
    buildClassifier in class RegSMO
    
    Parameters:
    
    instances - the data to work with
    
    Throws:
    
    java.lang.Exception - if something goes wrong

Class RegSMOImproved

Field Summary

Fields inherited from class keel.Algorithms.SVM.SMO.supportVector.RegSMO

Fields inherited from class keel.Algorithms.SVM.SMO.supportVector.RegOptimizer

Constructor Summary

Method Summary

Methods inherited from class keel.Algorithms.SVM.SMO.supportVector.RegSMO

Methods inherited from class keel.Algorithms.SVM.SMO.supportVector.RegOptimizer

Methods inherited from class java.lang.Object

Field Detail

I0

I0a

I0b

I1

I2

I3

m_I0

m_iSet

m_bUp

m_bLow

m_iUp

m_iLow

Constructor Detail

RegSMOImproved

Method Detail

globalInfo

getTechnicalInformation

listOptions

setOptions

getOptions

toleranceTipText

getTolerance

setTolerance

useVariant1TipText

isUseVariant1

setUseVariant1

takeStep

updateIndexSetFor

updateBoundaries

examineExample

init

optimize1

optimize2

wrapUp

buildClassifier