fr.lip6.classifier
Class EnhancedSMOSVM<T>

java.lang.Object
  fr.lip6.classifier.EnhancedSMOSVM<T>

Type Parameters:

T - Type de donnée de l'espace d'entrée.

All Implemented Interfaces:

Classifier<T>, java.io.Serializable

public class EnhancedSMOSVM<T>
extends java.lang.Object
implements Classifier<T>, java.io.Serializable

Classifieur SVM utilisant l'algorithme SMO de J. Platt avec l'implementation de weka.
N'importe quel type peut être traité, pourvu qu'un noyau Kernel Soit fournit.

Author:

dpicard

See Also:

Serialized Form

Field Summary

protected double	m_bLow The thresholds.
protected double	m_bUp The thresholds.
protected double[]	m_class The transformed class values.
protected java.util.ArrayList<T>	m_data The training data.
protected static double	m_Del Precision constant for updating sets
protected double[]	m_errors The current set of errors for all non-bound examples.
protected java.util.TreeSet<java.lang.Integer>	m_I0 {i: 0 < m_alpha[i] < C}
protected java.util.TreeSet<java.lang.Integer>	m_I1 {i: m_class[i] = 1, m_alpha[i] = 0}
protected java.util.TreeSet<java.lang.Integer>	m_I2 {i: m_class[i] = -1, m_alpha[i] =C}
protected java.util.TreeSet<java.lang.Integer>	m_I3 {i: m_class[i] = 1, m_alpha[i] = C}
protected java.util.TreeSet<java.lang.Integer>	m_I4 {i: m_class[i] = -1, m_alpha[i] = 0}
protected int	m_iLow The indices for m_bLow and m_bUp
protected int	m_iUp The indices for m_bLow and m_bUp
protected int[]	m_sparseIndices
protected double[]	m_sparseWeights Variables to hold weight vector in sparse form.
protected double	m_sumOfWeights Stores the weight of the training instances
protected java.util.TreeSet<java.lang.Integer>	m_supportVectors The set of support vectors
protected double[]	m_weights Weight vector for linear machine.

Constructor Summary

EnhancedSMOSVM(Kernel<T> k)
Constructeur passant le noyau servant à calculer la similarité entre les éléments de l'espace d'entrée.

Method Summary

protected boolean	examineExample(int i2) Examines instance.
double[]	getAlphas()
protected boolean	takeStep(int i1, int i2, double F2) Method solving for the Lagrange multipliers for two instances.
void	train(java.util.ArrayList<TrainingSample<T>> t) Replace the current training and train the classifier
void	train(TrainingSample<T> t) Add a single example to the current training set and train the classifier
double	valueOf(T e) Computes the category of the provided example

Methods inherited from class java.lang.Object

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

Field Detail

m_bLow

protected double m_bLow

The thresholds.

m_bUp

protected double m_bUp

The thresholds.

m_iLow

protected int m_iLow

The indices for m_bLow and m_bUp

m_iUp

protected int m_iUp

The indices for m_bLow and m_bUp

m_data

protected java.util.ArrayList<T> m_data

The training data.

m_weights

protected double[] m_weights

Weight vector for linear machine.

m_sparseWeights

protected double[] m_sparseWeights

Variables to hold weight vector in sparse form. (To reduce storage requirements.)

m_sparseIndices

protected int[] m_sparseIndices

m_class

protected double[] m_class

The transformed class values.

m_errors

protected double[] m_errors

The current set of errors for all non-bound examples.

m_I0

protected java.util.TreeSet<java.lang.Integer> m_I0

{i: 0 < m_alpha[i] < C}

m_I1

protected java.util.TreeSet<java.lang.Integer> m_I1

{i: m_class[i] = 1, m_alpha[i] = 0}

m_I2

protected java.util.TreeSet<java.lang.Integer> m_I2

{i: m_class[i] = -1, m_alpha[i] =C}

m_I3

protected java.util.TreeSet<java.lang.Integer> m_I3

{i: m_class[i] = 1, m_alpha[i] = C}

m_I4

protected java.util.TreeSet<java.lang.Integer> m_I4

{i: m_class[i] = -1, m_alpha[i] = 0}

m_supportVectors

protected java.util.TreeSet<java.lang.Integer> m_supportVectors

The set of support vectors

m_sumOfWeights

protected double m_sumOfWeights

Stores the weight of the training instances

m_Del

protected static double m_Del

Precision constant for updating sets

Constructor Detail

EnhancedSMOSVM

public EnhancedSMOSVM(Kernel<T> k)

Constructeur passant le noyau servant à calculer la similarité entre les éléments de l'espace d'entrée.

Parameters:

k - le noyau templatisé en

Method Detail

train

public void train(TrainingSample<T> t)

Description copied from interface: Classifier

Add a single example to the current training set and train the classifier

Specified by:

train in interface Classifier<T>

Parameters:

t - the training sample

train

public void train(java.util.ArrayList<TrainingSample<T>> t)

Description copied from interface: Classifier

Replace the current training and train the classifier

Specified by:

train in interface Classifier<T>

Parameters:

t - list of training samples

examineExample

protected boolean examineExample(int i2)

Examines instance.

Parameters:

i2 - index of instance to examine

Returns:

true if examination was successfull

Throws:

java.lang.Exception - if something goes wrong

takeStep

protected boolean takeStep(int i1,
                           int i2,
                           double F2)

Method solving for the Lagrange multipliers for two instances.

Parameters:

i1 - index of the first instance

i2 - index of the second instance

F2 -

Returns:

true if multipliers could be found

Throws:

java.lang.Exception - if something goes wrong

valueOf

public double valueOf(T e)

Description copied from interface: Classifier

Computes the category of the provided example

Specified by:

valueOf in interface Classifier<T>

Parameters:

e - example

Returns:

>0. if e belongs to the category, <0. if not.

getAlphas

public double[] getAlphas()