weka.knowledgeflow.steps

Class ASEvaluator

java.lang.Object

weka.knowledgeflow.steps.BaseStep

weka.knowledgeflow.steps.WekaAlgorithmWrapper

weka.knowledgeflow.steps.ASEvaluator

All Implemented Interfaces:

java.io.Serializable, BaseStepExtender, Step

@KFStep(name="ASEvaluator",
        category="AttSelection",
        toolTipText="Weka attribute selection evaluator wrapper",
        iconPath="",
        resourceIntensive=true)
public class ASEvaluator
extends WekaAlgorithmWrapper

Step that wraps a Weka attribute or subset evaluator. Handles training and test set connections. Requires an ASSearchStrategy step to be connected via an "info" connection. Will output both attribute selection results (via text connections) and transformed data (via outgoing train or test set connections). When processing multiple incoming training and test folds, the step can either output a cross-validation style summary over all the folds or individual attribute selection results for each fold.

See Also:

ASSearchStrategy, Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected AttributeSelection	m_eval Eval to use when performing a cross-validation and not outputting separate results for each fold
protected ASEvaluation	m_evaluatorTemplate The evaluator (attribute or subset) being used
protected boolean	m_isDoingXVal True if we are processing cross-validation folds to produce a summary over the folds (as opposed to producing separate results per fold).
protected boolean	m_isRanking Whether a ranking is being produced by the attribute selection
protected boolean	m_isReset True if we've been reset
protected java.util.Map<java.lang.Integer,java.lang.Integer>	m_numToSelectStore Holds the calculated number of attributes to select (may depend on thresholds) for each training fold
protected ASSearch	m_searchTemplate The search strategy being used (retrieved via an incoming "info" connection
protected java.util.Map<java.lang.Integer,int[]>	m_selectedAttsStore Holds selected attribute indices corresponding to training folds
protected java.util.concurrent.atomic.AtomicInteger	m_setCount Keeps count of the folds processed
protected java.util.Map<java.lang.Integer,AttributeTransformer>	m_transformerStore Holds the evaluator trained per fold in the case when it is a transformer (such as PCA)
protected boolean	m_treatXValFoldsSeparately Whether to output separate evaluation results for each fold of a xval or report the cross-validation summary
protected java.util.Map<java.lang.Integer,Instances>	m_waitingTestData Any test folds waiting to be processed (i.e. have their dimensionality reduced

Fields inherited from class weka.knowledgeflow.steps.WekaAlgorithmWrapper

m_defaultIconPath, m_defaultPackageIconPath, m_iconPath, m_wrappedAlgorithm

Fields inherited from class weka.knowledgeflow.steps.BaseStep

m_stepIsResourceIntensive, m_stepManager, m_stepName

Constructor Summary

Constructors

Constructor and Description
ASEvaluator()

Method Summary

Modifier and Type	Method and Description
protected void	applyFiltering(java.lang.String connType, int[] selectedAtts, Instances data, java.lang.Integer setNum, java.lang.Integer maxSetNum) Apply a filter to reduce the dimensionality of the supplied data.
protected void	checkTestFiltering(Instances test, java.lang.Integer setNum, java.lang.Integer maxSetNum) Check to see if there is a waiting set of selected attributes that can be used to reduce the dimensionality of the supplied test set
java.lang.String	getCustomEditorForStep() Get the class name of the custom editor for this step
ASEvaluation	getEvaluator() Get the evaluator wrapped by this step
java.util.List<java.lang.String>	getIncomingConnectionTypes() Get incoming connections accepted given the current state of the step
java.util.List<java.lang.String>	getOutgoingConnectionTypes() Get a list of output connections that can be produced given the current state of the step
boolean	getTreatXValFoldsSeparately() Get whether to output separate results for each fold of a cross-validation, rather than averaging over folds.
java.lang.Class	getWrappedAlgorithmClass() Get the class of Weka algorithm wrapped by this wrapper
protected void	outputTextData(java.lang.String text, java.lang.Integer setNum) Output Data to outgoing text connections
void	processIncoming(Data data) Process an incoming Data object
protected void	processNonXVal(Instances train, Instances test, java.lang.Integer setNum, java.lang.Integer maxSetNum) Handles processing for single train sets, single train/test pairs or cross-validation folds when results are output for each separately
protected void	processXVal(Instances train, Instances test, java.lang.Integer setNum, java.lang.Integer maxSetNum) Handles processing in the cross-validation case when results are to be accumulated/averaged over the folds
void	setEvaluator(ASEvaluation eval) Set the evaluator to wrap (just calls setWrappedAlgorithm)
void	setTreatXValFoldsSeparately(boolean treatSeparately) Set whether to output separate results for each fold of a cross-validation, rather than averaging over folds.
void	setWrappedAlgorithm(java.lang.Object algo) Set an instance of the wrapped algorithm to use
void	stepInit() Initialize at the start of a run
protected void	updateXValStats(Instances train, ASEvaluation evaluator, ASSearch search, int[] selectedAtts) Updates stats in the cross-validation case

Methods inherited from class weka.knowledgeflow.steps.WekaAlgorithmWrapper

getDefaultIconPath, getDefaultPackageLevelIconPath, getIconPath, getWrappedAlgorithm, globalInfo

Methods inherited from class weka.knowledgeflow.steps.BaseStep

environmentSubstitute, getDefaultSettings, getInteractiveViewers, getInteractiveViewersImpls, getName, getStepManager, isResourceIntensive, isStopRequested, outputStructureForConnectionType, setName, setStepIsResourceIntensive, setStepManager, setStepMustRunSingleThreaded, start, stepMustRunSingleThreaded, stop

Methods inherited from class java.lang.Object

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

Field Detail

m_evaluatorTemplate

protected ASEvaluation m_evaluatorTemplate

The evaluator (attribute or subset) being used

m_searchTemplate

protected ASSearch m_searchTemplate

The search strategy being used (retrieved via an incoming "info" connection

m_waitingTestData

protected java.util.Map<java.lang.Integer,Instances> m_waitingTestData

Any test folds waiting to be processed (i.e. have their dimensionality reduced

m_selectedAttsStore

protected java.util.Map<java.lang.Integer,int[]> m_selectedAttsStore

Holds selected attribute indices corresponding to training folds

m_numToSelectStore

protected java.util.Map<java.lang.Integer,java.lang.Integer> m_numToSelectStore

Holds the calculated number of attributes to select (may depend on thresholds) for each training fold

m_transformerStore

protected java.util.Map<java.lang.Integer,AttributeTransformer> m_transformerStore

Holds the evaluator trained per fold in the case when it is a transformer (such as PCA)

m_isReset

protected boolean m_isReset

True if we've been reset

m_isDoingXVal

protected boolean m_isDoingXVal

True if we are processing cross-validation folds to produce a summary over the folds (as opposed to producing separate results per fold).

m_setCount

protected java.util.concurrent.atomic.AtomicInteger m_setCount

Keeps count of the folds processed

m_treatXValFoldsSeparately

protected boolean m_treatXValFoldsSeparately

Whether to output separate evaluation results for each fold of a xval or report the cross-validation summary

m_isRanking

protected boolean m_isRanking

Whether a ranking is being produced by the attribute selection

m_eval

protected AttributeSelection m_eval

Eval to use when performing a cross-validation and not outputting separate results for each fold

Constructor Detail

ASEvaluator

public ASEvaluator()

Method Detail

getWrappedAlgorithmClass

public java.lang.Class getWrappedAlgorithmClass()

Get the class of Weka algorithm wrapped by this wrapper

Specified by:

getWrappedAlgorithmClass in class WekaAlgorithmWrapper

Returns:

the wrapped algorithm class

setWrappedAlgorithm

public void setWrappedAlgorithm(java.lang.Object algo)

Set an instance of the wrapped algorithm to use

Overrides:

setWrappedAlgorithm in class WekaAlgorithmWrapper

Parameters:

algo - the algorithm to use

getEvaluator

public ASEvaluation getEvaluator()

Get the evaluator wrapped by this step

Returns:

the attribute or subset evaluator wrapped by this step

setEvaluator

@ProgrammaticProperty
public void setEvaluator(ASEvaluation eval)

Set the evaluator to wrap (just calls setWrappedAlgorithm)

Parameters:

eval - the evaluator to use

setTreatXValFoldsSeparately

@OptionMetadata(displayName="Treat x-val folds separately",
                description="Output separate attribute selection results for each fold of a cross-validation (rather than averaging across folds)")
public void setTreatXValFoldsSeparately(boolean treatSeparately)

Set whether to output separate results for each fold of a cross-validation, rather than averaging over folds.

Parameters:

treatSeparately - true if each fold will have results output

getTreatXValFoldsSeparately

public boolean getTreatXValFoldsSeparately()

Get whether to output separate results for each fold of a cross-validation, rather than averaging over folds.

Returns:

true if each fold will have results output

stepInit

public void stepInit()
              throws WekaException

Initialize at the start of a run

Throws:

WekaException - if there is an illegal configuration (i.e. Ranker search with subset evaluator or regular search with attribute evaluator

processIncoming

public void processIncoming(Data data)
                     throws WekaException

Process an incoming Data object

Specified by:

processIncoming in interface BaseStepExtender

Specified by:

processIncoming in interface Step

Overrides:

processIncoming in class BaseStep

Parameters:

data - the data object to process

Throws:

WekaException - if a problem occurs

outputTextData

protected void outputTextData(java.lang.String text,
                              java.lang.Integer setNum)
                       throws WekaException

Output Data to outgoing text connections

Parameters:

text - the text to output

setNum - the fold/set number that this text is associated with

Throws:

WekaException - if a problem occurs

processNonXVal

protected void processNonXVal(Instances train,
                              Instances test,
                              java.lang.Integer setNum,
                              java.lang.Integer maxSetNum)
                       throws WekaException

Handles processing for single train sets, single train/test pairs or cross-validation folds when results are output for each separately

Parameters:

train - the training data

test - the test data

setNum - the set number for the training/test data

maxSetNum - the maximum set number

Throws:

WekaException - if a problem occurs

processXVal

protected void processXVal(Instances train,
                           Instances test,
                           java.lang.Integer setNum,
                           java.lang.Integer maxSetNum)
                    throws WekaException

Handles processing in the cross-validation case when results are to be accumulated/averaged over the folds

Parameters:

train - the training data

test - the test data

setNum - the set number of this train/test pair

maxSetNum - the maximum set number

Throws:

WekaException - if a problem occurs

checkTestFiltering

protected void checkTestFiltering(Instances test,
                                  java.lang.Integer setNum,
                                  java.lang.Integer maxSetNum)
                           throws WekaException

Check to see if there is a waiting set of selected attributes that can be used to reduce the dimensionality of the supplied test set

Parameters:

test - the test set to potentially filter

setNum - the set number of the test set

maxSetNum - the maximum set number

Throws:

WekaException - if a problem occurs

applyFiltering

protected void applyFiltering(java.lang.String connType,
                              int[] selectedAtts,
                              Instances data,
                              java.lang.Integer setNum,
                              java.lang.Integer maxSetNum)
                       throws WekaException

Apply a filter to reduce the dimensionality of the supplied data. Outputs the reduced data to downstream steps on the given connection type

Parameters:

connType - the connection type to output on

selectedAtts - selected attribute indices to use when filtering

data - the instances to filter

setNum - the set number of the instances

maxSetNum - the maximum set number

Throws:

WekaException - if a problem occurs

updateXValStats

protected void updateXValStats(Instances train,
                               ASEvaluation evaluator,
                               ASSearch search,
                               int[] selectedAtts)
                        throws java.lang.Exception

Updates stats in the cross-validation case

Parameters:

train - the training data processed

evaluator - the evaluator used

search - the search strategy

selectedAtts - the attributes selected on this training data

Throws:

java.lang.Exception - if a problem occurs

getIncomingConnectionTypes

public java.util.List<java.lang.String> getIncomingConnectionTypes()

Get incoming connections accepted given the current state of the step

Returns:

a list of acceptable incoming connections

getOutgoingConnectionTypes

public java.util.List<java.lang.String> getOutgoingConnectionTypes()

Get a list of output connections that can be produced given the current state of the step

Returns:

a list of output connections

getCustomEditorForStep

public java.lang.String getCustomEditorForStep()

Get the class name of the custom editor for this step

Specified by:

getCustomEditorForStep in interface Step

Overrides:

getCustomEditorForStep in class BaseStep

Returns:

the class name of the custom editor for this step