org.aksw.limes.core.ml.algorithm.eagle.core

Class ExpressionFitnessFunction

java.lang.Object

org.jgap.gp.GPFitnessFunction

org.aksw.limes.core.ml.algorithm.eagle.core.IGPFitnessFunction

org.aksw.limes.core.ml.algorithm.eagle.core.ExpressionFitnessFunction

All Implemented Interfaces:

Serializable, IFitnessFunction

public class ExpressionFitnessFunction
extends IGPFitnessFunction

Implementation of our custom FitnessFunction. As we're using the DeltaFitnessEvaluator higher fitness values mean the individual is less fit! ReImplementation on ExcutionEngine

FIXME fix QualityMeasures to work on Mappings!

Author:

Klaus Lyko, Tommaso Soru (tsoru@informatik.uni-leipzig.de)

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected double	crossProduct
static String	fScore
protected LinkSpecGeneticLearnerConfig	m_config
protected IQualitativeMeasure	measure
static String	precision
static String	recall
protected AMapping	reference Fragment of optimal Mapping used during evolution.
protected ACache	sC
protected ACache	tC
protected ACache	trimmedSourceCache
protected ACache	trimmedTargetCache

Fields inherited from class org.jgap.gp.GPFitnessFunction

MAX_FITNESS_VALUE, NO_FITNESS_VALUE

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	ExpressionFitnessFunction(LinkSpecGeneticLearnerConfig a_config) Needed for subclasses.

Method Summary

Modifier and Type	Method and Description
void	addToReference(AMapping m) Method to add instances to reference?.
double	calculateRawFitness(org.jgap.gp.IGPProgram p) Calculates the fitness of the given GPProgram p.
double	calculateRawMeasure(org.jgap.gp.IGPProgram p)
void	destroy()
protected double	evaluate(org.jgap.gp.IGPProgram a_subject)
void	fillCachesIncrementally(AMapping matches)
static ExpressionFitnessFunction	getInstance(LinkSpecGeneticLearnerConfig a_config, IQualitativeMeasure measure, AMapping reference)
AMapping	getMapping(ACache sourceCache, ACache targetCache, LinkSpecification spec) To calculate Mappings based on LS
LinkSpecification	getMetric(org.jgap.gp.IGPProgram p)
AMapping	getReferenceMapping()
ACache	getSourceCache() Get full caches used by this learner.
ACache	getTargetCache() Get full caches used by this learner.
void	setCaches(ACache sC, ACache tC)
void	setReferenceMapping(AMapping referenceData) As we assume referenceData only holding matches.
void	setUseFullCaches(boolean value) If we want to use Full Caches instead.
void	trimKnowledgeBases(AMapping trainingData) Method to scale down caches according to given training data.

Methods inherited from class org.jgap.gp.GPFitnessFunction

getFitnessValue, getLastComputedFitnessValue, GPFitnessFunction

Methods inherited from class java.lang.Object

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

Field Detail

fScore

public static final String fScore

See Also:

Constant Field Values

recall

public static final String recall

See Also:

Constant Field Values

precision

public static final String precision

See Also:

Constant Field Values

m_config

protected LinkSpecGeneticLearnerConfig m_config

reference

protected AMapping reference

Fragment of optimal Mapping used during evolution. Note that it should only hold matches!

sC

protected ACache sC

tC

protected ACache tC

trimmedSourceCache

protected ACache trimmedSourceCache

trimmedTargetCache

protected ACache trimmedTargetCache

measure

protected IQualitativeMeasure measure

crossProduct

protected double crossProduct

Constructor Detail

ExpressionFitnessFunction

protected ExpressionFitnessFunction(LinkSpecGeneticLearnerConfig a_config)

Needed for subclasses.

Method Detail

getInstance

public static ExpressionFitnessFunction getInstance(LinkSpecGeneticLearnerConfig a_config,
                                                    IQualitativeMeasure measure,
                                                    AMapping reference)

evaluate

protected double evaluate(org.jgap.gp.IGPProgram a_subject)

Specified by:

evaluate in class org.jgap.gp.GPFitnessFunction

calculateRawFitness

public double calculateRawFitness(org.jgap.gp.IGPProgram p)

Calculates the fitness of the given GPProgram p. This is done as follows: 1.1 get the Metric Expression by executing (as object) the first chromosome. 1.2 get the mapping links with these settings 2. Calculate either recall, precision or f-score and return 1 minus it. The evolution could produce non wanted individuals especially those who compare properties which are not part of the PropertyMapping (set in the ExpressionConfiguration). In such cases setp 1.1 throws an IllegalStateException. As a result the fitness value would be set to rather bad one.

Parameters:

p - GPProgram fitness is calculated for.

Returns:

Double value. The closer to 0 the better.

calculateRawMeasure

public double calculateRawMeasure(org.jgap.gp.IGPProgram p)

getSourceCache

public ACache getSourceCache()

Get full caches used by this learner.

Returns:

HybridCache of the source endpoint.

getTargetCache

public ACache getTargetCache()

Get full caches used by this learner.

Returns:

HybridCache of the target endpoint.

destroy

public void destroy()

getMapping

public AMapping getMapping(ACache sourceCache,
                           ACache targetCache,
                           LinkSpecification spec)

Description copied from interface: IFitnessFunction

To calculate Mappings based on LS

Returns:

trimKnowledgeBases

public void trimKnowledgeBases(AMapping trainingData)

Method to scale down caches according to given training data.

Parameters:

trainingData - Mapping holding data instances a user has evaluated. That may include non-matches.

setUseFullCaches

public void setUseFullCaches(boolean value)

If we want to use Full Caches instead.

Parameters:

value -

getMetric

public LinkSpecification getMetric(org.jgap.gp.IGPProgram p)

getReferenceMapping

public AMapping getReferenceMapping()

setReferenceMapping

public void setReferenceMapping(AMapping referenceData)

As we assume referenceData only holding matches. Learner may have to set it separately.

Parameters:

referenceData - A Mapping holding all matches.

addToReference

public void addToReference(AMapping m)

Method to add instances to reference?.

Specified by:

addToReference in class IGPFitnessFunction

Parameters:

m - Mapping of matches, designated as such by an oracle.

fillCachesIncrementally

public void fillCachesIncrementally(AMapping matches)

Specified by:

fillCachesIncrementally in class IGPFitnessFunction

setCaches

public void setCaches(ACache sC,
                      ACache tC)