weka.attributeSelection
Class AttributeSelection

java.lang.Object
  weka.attributeSelection.AttributeSelection

All Implemented Interfaces:

java.io.Serializable

public class AttributeSelection

extends java.lang.Object

implements java.io.Serializable

Attribute selection class. Takes the name of a search class and an evaluation class on the command line.

Valid options are:

-h
Display help.

-I
Specify the training arff file.

-C
The index of the attribute to use as the class.

-S

The full class name of the search method followed by search method options (if any).
Eg. -S "weka.attributeSelection.BestFirst -N 10"

-X
Perform a cross validation.

-N
Specify a random number seed. Use in conjuction with -X. (Default = 1).

------------------------------------------------------------------------

Example usage as the main of an attribute evaluator (called FunkyEvaluator):

 public static void main(String [] args) {
   try {
     ASEvaluator eval = new FunkyEvaluator();
     System.out.println(SelectAttributes(Evaluator, args));
   } catch (Exception e) {
     System.err.println(e.getMessage());
   }
 }
  

------------------------------------------------------------------------

Version:

$Revision: 1.31 $

Author:

Mark Hall (mhall@cs.waikato.ac.nz)

See Also:

Serialized Form

Field Summary

private ASEvaluation	m_ASEvaluator the attribute/subset evaluator
private Remove	m_attributeFilter the attribute filter for processing instances with respect to the most recent feature selection run
private double[][]	m_attributeRanking the attribute indexes and associated merits if a ranking is produced
private boolean	m_doRank rank features (if allowed by the search method)
private boolean	m_doXval do cross validation
private int	m_numFolds the number of folds to use for cross validation
private int	m_numToSelect number of attributes requested from ranked results
private double[][]	m_rankResults hold statistics for repeated feature selection, such as under cross validation
private ASSearch	m_searchMethod the search method
private int	m_seed seed used to randomly shuffle instances for cross validation
private int[]	m_selectedAttributeSet the selected attributes
private java.lang.StringBuffer	m_selectionResults holds a string describing the results of the attribute selection
private double[]	m_subsetResults
private double	m_threshold cutoff value by which to select attributes for ranked results
private Instances	m_trainInstances the instances to select attributes from
private AttributeTransformer	m_transformer if a feature selection run involves an attribute transformer
private int	m_trials

Constructor Summary

AttributeSelection()
constructor.

Method Summary

java.lang.String	CrossValidateAttributes() Perform a cross validation for attribute selection.
java.lang.String	CVResultsString() returns a string summarizing the results of repeated attribute selection runs on splits of a dataset.
static void	main(java.lang.String[] args) Main method for testing this class.
private static java.lang.String	makeOptionString(ASEvaluation ASEvaluator, ASSearch searchMethod) Make up the help string giving all the command line options
int	numberAttributesSelected() Return the number of attributes selected from the most recent run of attribute selection
private java.lang.String	printSelectionResults() Assembles a text description of the attribute selection results.
double[][]	rankedAttributes() get the final ranking of the attributes.
Instance	reduceDimensionality(Instance in) reduce the dimensionality of a single instance to include only those attributes chosen by the last run of attribute selection.
Instances	reduceDimensionality(Instances in) reduce the dimensionality of a set of instances to include only those attributes chosen by the last run of attribute selection.
static java.lang.String	SelectAttributes(ASEvaluation ASEvaluator, java.lang.String[] options) Perform attribute selection with a particular evaluator and a set of options specifying search method and input file etc.
static java.lang.String	SelectAttributes(ASEvaluation ASEvaluator, java.lang.String[] options, Instances train) Perform attribute selection with a particular evaluator and a set of options specifying search method and options for the search method and evaluator.
void	SelectAttributes(Instances data) Perform attribute selection on the supplied training instances.
void	selectAttributesCVSplit(Instances split) Select attributes for a split of the data.
int[]	selectedAttributes() get the final selected set of attributes.
void	setEvaluator(ASEvaluation evaluator) set the attribute/subset evaluator
void	setFolds(int folds) set the number of folds for cross validation
void	setRanking(boolean r) produce a ranking (if possible with the set search and evaluator)
void	setSearch(ASSearch search) set the search method
void	setSeed(int s) set the seed for use in cross validation
void	setThreshold(double t) set the threshold by which to select features from a ranked list
void	setXval(boolean x) do a cross validation
java.lang.String	toResultsString() get a description of the attribute selection

Methods inherited from class java.lang.Object

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

Field Detail

m_trainInstances

private Instances m_trainInstances

the instances to select attributes from

m_ASEvaluator

private ASEvaluation m_ASEvaluator

the attribute/subset evaluator

m_searchMethod

private ASSearch m_searchMethod

the search method

m_numFolds

private int m_numFolds

the number of folds to use for cross validation

m_selectionResults

private java.lang.StringBuffer m_selectionResults

holds a string describing the results of the attribute selection

m_doRank

private boolean m_doRank

rank features (if allowed by the search method)

m_doXval

private boolean m_doXval

do cross validation

m_seed

private int m_seed

seed used to randomly shuffle instances for cross validation

m_threshold

private double m_threshold

cutoff value by which to select attributes for ranked results

m_numToSelect

private int m_numToSelect

number of attributes requested from ranked results

m_selectedAttributeSet

private int[] m_selectedAttributeSet

the selected attributes

m_attributeRanking

private double[][] m_attributeRanking

the attribute indexes and associated merits if a ranking is produced

m_transformer

private AttributeTransformer m_transformer

if a feature selection run involves an attribute transformer

m_attributeFilter

private Remove m_attributeFilter

the attribute filter for processing instances with respect to the most recent feature selection run

m_rankResults

private double[][] m_rankResults

hold statistics for repeated feature selection, such as under cross validation

m_subsetResults

private double[] m_subsetResults

m_trials

private int m_trials

Constructor Detail

AttributeSelection

public AttributeSelection()

constructor. Sets defaults for each member varaible. Default attribute evaluator is CfsSubsetEval; default search method is BestFirst.

Method Detail

numberAttributesSelected

public int numberAttributesSelected()
                             throws java.lang.Exception

Return the number of attributes selected from the most recent run of attribute selection

Returns:

the number of attributes selected

Throws:

java.lang.Exception

selectedAttributes

public int[] selectedAttributes()
                         throws java.lang.Exception

get the final selected set of attributes.

Returns:

an array of attribute indexes

Throws:

java.lang.Exception - if attribute selection has not been performed yet

rankedAttributes

public double[][] rankedAttributes()
                            throws java.lang.Exception

get the final ranking of the attributes.

Returns:

a two dimensional array of ranked attribute indexes and their associated merit scores as doubles.

Throws:

java.lang.Exception - if a ranking has not been produced

setEvaluator

public void setEvaluator(ASEvaluation evaluator)

set the attribute/subset evaluator

Parameters:

evaluator - the evaluator to use

setSearch

public void setSearch(ASSearch search)

set the search method

Parameters:

search - the search method to use

setFolds

public void setFolds(int folds)

set the number of folds for cross validation

Parameters:

folds - the number of folds

setRanking

public void setRanking(boolean r)

produce a ranking (if possible with the set search and evaluator)

Parameters:

r - true if a ranking is to be produced

setXval

public void setXval(boolean x)

do a cross validation

Parameters:

x - true if a cross validation is to be performed

setSeed

public void setSeed(int s)

set the seed for use in cross validation

Parameters:

s - the seed

setThreshold

public void setThreshold(double t)

set the threshold by which to select features from a ranked list

Parameters:

t - the threshold

toResultsString

public java.lang.String toResultsString()

get a description of the attribute selection

Returns:

a String describing the results of attribute selection

reduceDimensionality

public Instances reduceDimensionality(Instances in)
                               throws java.lang.Exception

reduce the dimensionality of a set of instances to include only those attributes chosen by the last run of attribute selection.

Parameters:

in - the instances to be reduced

Returns:

a dimensionality reduced set of instances

Throws:

java.lang.Exception - if the instances can't be reduced

reduceDimensionality

public Instance reduceDimensionality(Instance in)
                              throws java.lang.Exception

reduce the dimensionality of a single instance to include only those attributes chosen by the last run of attribute selection.

Parameters:

in - the instance to be reduced

Returns:

a dimensionality reduced instance

Throws:

java.lang.Exception - if the instance can't be reduced

SelectAttributes

public static java.lang.String SelectAttributes(ASEvaluation ASEvaluator,
                                                java.lang.String[] options)
                                         throws java.lang.Exception

Perform attribute selection with a particular evaluator and a set of options specifying search method and input file etc.

Parameters:

ASEvaluator - an evaluator object

options - an array of options, not only for the evaluator but also the search method (if any) and an input data file

Returns:

the results of attribute selection as a String

Throws:

java.lang.Exception - if no training file is set

CVResultsString

public java.lang.String CVResultsString()
                                 throws java.lang.Exception

returns a string summarizing the results of repeated attribute selection runs on splits of a dataset.

Returns:

a summary of attribute selection results

Throws:

java.lang.Exception - if no attribute selection has been performed.

selectAttributesCVSplit

public void selectAttributesCVSplit(Instances split)
                             throws java.lang.Exception

Select attributes for a split of the data. Calling this function updates the statistics on attribute selection. CVResultsString() returns a string summarizing the results of repeated calls to this function. Assumes that splits are from the same dataset--- ie. have the same number and types of attributes as previous splits.

Parameters:

split - the instances to select attributes from

Throws:

java.lang.Exception - if an error occurs

CrossValidateAttributes

public java.lang.String CrossValidateAttributes()
                                         throws java.lang.Exception

Perform a cross validation for attribute selection. With subset evaluators the number of times each attribute is selected over the cross validation is reported. For attribute evaluators, the average merit and average ranking + std deviation is reported for each attribute.

Returns:

the results of cross validation as a String

Throws:

java.lang.Exception - if an error occurs during cross validation

SelectAttributes

public void SelectAttributes(Instances data)
                      throws java.lang.Exception

Perform attribute selection on the supplied training instances.

Parameters:

data - the instances to select attributes from

Throws:

java.lang.Exception - if there is a problem during selection

SelectAttributes

public static java.lang.String SelectAttributes(ASEvaluation ASEvaluator,
                                                java.lang.String[] options,
                                                Instances train)
                                         throws java.lang.Exception

Perform attribute selection with a particular evaluator and a set of options specifying search method and options for the search method and evaluator.

Parameters:

ASEvaluator - an evaluator object

options - an array of options, not only for the evaluator but also the search method (if any) and an input data file

train - the input instances

Returns:

the results of attribute selection as a String

Throws:

java.lang.Exception - if incorrect options are supplied

printSelectionResults

private java.lang.String printSelectionResults()

Assembles a text description of the attribute selection results.

Returns:

a string describing the results of attribute selection.

makeOptionString

private static java.lang.String makeOptionString(ASEvaluation ASEvaluator,
                                                 ASSearch searchMethod)
                                          throws java.lang.Exception

Make up the help string giving all the command line options

Parameters:

ASEvaluator - the attribute evaluator to include options for

searchMethod - the search method to include options for

Returns:

a string detailing the valid command line options

Throws:

java.lang.Exception

main

public static void main(java.lang.String[] args)

Main method for testing this class.

Parameters:

args - the options