dragon.ir.classification.featureselection
Class AbstractFeatureSelector

java.lang.Object
  |
  +--dragon.ir.classification.featureselection.AbstractFeatureSelector

All Implemented Interfaces:

FeatureSelector, java.io.Serializable

Direct Known Subclasses:

ChiFeatureSelector, DocFrequencySelector, InfoGainFeatureSelector, MutualInfoFeatureSelector, NullFeatureSelector

public abstract class AbstractFeatureSelector

extends java.lang.Object

implements FeatureSelector, java.io.Serializable

Abstract function class for feature selection

Copyright: Copyright (c) 2005

Company: IST, Drexel University

Version:

1.0

Author:

Davis Zhou

See Also:

Serialized Form

Field Summary

protected int[]	featureMap
protected int	selectedFeatureNum

Constructor Summary

AbstractFeatureSelector()

Method Summary

protected DoubleVector	getClassPrior(DocClassSet docSet)
int	getSelectedFeatureNum()
protected abstract int[]	getSelectedFeatures(IndexReader indexReader, DocClassSet trainingSet)
protected abstract int[]	getSelectedFeatures(SparseMatrix doctermMatrix, DocClassSet trainingSet)
protected IntDenseMatrix	getTermDistribution(IndexReader indexReader, DocClassSet trainingSet)
protected IntDenseMatrix	getTermDistribution(SparseMatrix doctermMatrix, DocClassSet trainingSet)
protected int[]	getTermDocFrequency(SparseMatrix matrix, DocClassSet trainingSet)
boolean	isSelected(int originalFeatureIndex)
int	map(int originalFeatureIndex) Map the old feature index to the index in the new feature space.
void	setSelectedFeatures(int[] selectedFeatures) Manually set selected features.
void	train(IndexReader indexReader, DocClassSet trainingSet) This method chooses a subset of features for text classification
void	train(SparseMatrix doctermMatrix, DocClassSet trainingSet) This method chooses a subset of features for text classification.

Methods inherited from class java.lang.Object

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

Field Detail

featureMap

protected int[] featureMap

selectedFeatureNum

protected int selectedFeatureNum

Constructor Detail

AbstractFeatureSelector

public AbstractFeatureSelector()

Method Detail

getSelectedFeatures

protected abstract int[] getSelectedFeatures(IndexReader indexReader,
                                             DocClassSet trainingSet)

getSelectedFeatures

protected abstract int[] getSelectedFeatures(SparseMatrix doctermMatrix,
                                             DocClassSet trainingSet)

train

public void train(IndexReader indexReader,
                  DocClassSet trainingSet)

Description copied from interface: FeatureSelector

This method chooses a subset of features for text classification

Specified by:

train in interface FeatureSelector

Parameters:

indexReader - the index reader a classifer is working on

trainingSet - the labeled training document set

train

public void train(SparseMatrix doctermMatrix,
                  DocClassSet trainingSet)

Description copied from interface: FeatureSelector

This method chooses a subset of features for text classification. Some implmentations such as DocFrequencySelector and InfoGainFeatureSelector do not support this method.

Specified by:

train in interface FeatureSelector

Parameters:

doctermMatrix - the document-term matrix a classifer is working on

trainingSet - the labeled training document set

setSelectedFeatures

public void setSelectedFeatures(int[] selectedFeatures)

Description copied from interface: FeatureSelector

Manually set selected features. Usually used in testing stage.

Specified by:

setSelectedFeatures in interface FeatureSelector

Parameters:

selectedFeatures - each elements contains the index of the selected feature in the old feature space. The selected feature must be in the ascending order in the input array.

isSelected

public boolean isSelected(int originalFeatureIndex)

Specified by:

isSelected in interface FeatureSelector

Parameters:

originalFeatureIndex - the index of the feature in the old feature space

Returns:

true if the given feature is selected for text classification

map

public int map(int originalFeatureIndex)

Description copied from interface: FeatureSelector

Map the old feature index to the index in the new feature space.

Specified by:

map in interface FeatureSelector

Parameters:

originalFeatureIndex - the index of the feature before feature selection

Returns:

the index of the feature the new space. If the feature is not selected, it will return -1.

getSelectedFeatureNum

public int getSelectedFeatureNum()

Specified by:

getSelectedFeatureNum in interface FeatureSelector

Returns:

the number of selected features

getClassPrior

protected DoubleVector getClassPrior(DocClassSet docSet)

getTermDocFrequency

protected int[] getTermDocFrequency(SparseMatrix matrix,
                                    DocClassSet trainingSet)

getTermDistribution

protected IntDenseMatrix getTermDistribution(IndexReader indexReader,
                                             DocClassSet trainingSet)

getTermDistribution

protected IntDenseMatrix getTermDistribution(SparseMatrix doctermMatrix,
                                             DocClassSet trainingSet)