org.aksw.limes.core.ml.algorithm.wombat

Class AWombat

java.lang.Object

org.aksw.limes.core.ml.algorithm.ACoreMLAlgorithm

org.aksw.limes.core.ml.algorithm.wombat.AWombat

Direct Known Subclasses:

WombatComplete, WombatSimple

public abstract class AWombat
extends ACoreMLAlgorithm

This class uses Least General Generalization (LGG) to learn Link Specifications (LS)

Version:

Jun 7, 2016

Author:

Mohamed Sherif (sherif@informatik.uni-leipzig.de)

Field Summary

Fields

Modifier and Type	Field and Description
protected boolean	isUnsupervised
static String	PARAMETER_ATOMIC_MEASURES
static String	PARAMETER_CHILDREN_PENALTY_WEIGHT
static String	PARAMETER_COMPLEXITY_PENALTY_WEIGHT
static String	PARAMETER_EXECUTION_TIME_IN_MINUTES
static String	PARAMETER_MAX_FITNESS_THRESHOLD
static String	PARAMETER_MAX_ITERATION_TIME_IN_MINUTES
static String	PARAMETER_MAX_ITERATIONS_NUMBER
static String	PARAMETER_MAX_REFINEMENT_TREE_SIZE
static String	PARAMETER_MIN_PROPERTY_COVERAGE
static String	PARAMETER_OVERALL_PENALTY_WEIGHT
static String	PARAMETER_PROPERTY_LEARNING_RATE
static String	PARAMETER_SAVE_MAPPING
static String	PARAMETER_VERBOSE
protected PseudoFMeasure	pseudoFMeasure
protected Tree<RefinementNode>	refinementTreeRoot
protected Map<String,Double>	sourcePropertiesCoverageMap
protected ACache	sourceSample
static List<String>	sourceUris
protected String	sourceVariable
protected Map<String,Double>	targetPropertiesCoverageMap
protected ACache	targetSample
static List<String>	targetUris
protected String	targetVariable
protected AMapping	trainingData
protected Set<String>	wombatParameterNames

Fields inherited from class org.aksw.limes.core.ml.algorithm.ACoreMLAlgorithm

configuration, learningParameters, sourceCache, targetCache

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	AWombat()

Method Summary

Modifier and Type	Method and Description
protected double	computePenalty(Tree<RefinementNode> promesyChild)
protected RefinementNode	createNode(AMapping mapping, String metricExpr) Create new RefinementNode using either real or pseudo-F-Measure
protected RefinementNode	createNode(String metricExpr)
AMapping	executeAtomicMeasure(String sourceProperty, String targetProperty, String measure, double threshold)
protected void	fillSampleSourceTargetCaches(AMapping sample)
protected ExtendedClassifier	findInitialClassifier(String sourceProperty, String targetProperty, String measure) Computes the atomic classifiers by finding the highest possible F-measure achievable on a given property pair
protected List<ExtendedClassifier>	findInitialClassifiers()
protected double	fMeasure(AMapping predictions) calculate either a real or a pseudo-F-Measure
protected Set<String>	getAtomicMeasures()
protected double	getChildrenPenaltyWeight()
protected double	getComplexityPenaltyWeight()
protected int	getExcutionTimeInMinutes()
protected int	getIterationTimeInMinutes()
protected AMapping	getMapingOfMetricExpression(String metricExpression) Looks first for the input metricExpression in the already constructed tree, if found the corresponding mapping is returned.
protected AMapping	getMapingOfMetricFromTree(String metricExpression, Tree<RefinementNode> r)
protected double	getMaxFitnessThreshold()
protected int	getMaxIterationNumber()
protected int	getMaxRefinmentTreeSize()
protected double	getMinPropertyCoverage()
protected Tree<RefinementNode>	getMostPromisingNode(Tree<RefinementNode> r, double penaltyWeight) Get the most promising node as the node with the best F-score
protected double	getOverAllPenaltyWeight()
protected AMapping	getPredictions(LinkSpecification ls, ACache sCache, ACache tCache) get mapping from source cache to target cache using metricExpression
protected double	getPropertyLearningRate()
PseudoFMeasure	getPseudoFMeasure()
AMapping	getTrainingData()
protected void	init(List<LearningParameter> lp, ACache sourceCache, ACache targetCache) Initialize the core ML algorithm.
boolean	isUnsupervised()
protected boolean	isVerbose()
protected double	precision(AMapping predictions) calculate either a real or a pseudo-Precision
protected double	recall(AMapping predictions) calculate either a real or a pseudo-Recall
void	setDefaultParameters() Set default ACoreMLAlgorithm parameters values

Methods inherited from class org.aksw.limes.core.ml.algorithm.ACoreMLAlgorithm

activeLearn, activeLearn, getConfiguration, getName, getNextExamples, getParameter, getParameters, getSourceCache, getTargetCache, learn, learn, predict, setConfiguration, setParameter, supports

Methods inherited from class java.lang.Object

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

Field Detail

PARAMETER_MAX_REFINEMENT_TREE_SIZE

public static final String PARAMETER_MAX_REFINEMENT_TREE_SIZE

See Also:

Constant Field Values

PARAMETER_MAX_ITERATIONS_NUMBER

public static final String PARAMETER_MAX_ITERATIONS_NUMBER

See Also:

Constant Field Values

PARAMETER_MAX_ITERATION_TIME_IN_MINUTES

public static final String PARAMETER_MAX_ITERATION_TIME_IN_MINUTES

See Also:

Constant Field Values

PARAMETER_EXECUTION_TIME_IN_MINUTES

public static final String PARAMETER_EXECUTION_TIME_IN_MINUTES

See Also:

Constant Field Values

PARAMETER_MAX_FITNESS_THRESHOLD

public static final String PARAMETER_MAX_FITNESS_THRESHOLD

See Also:

Constant Field Values

PARAMETER_MIN_PROPERTY_COVERAGE

public static final String PARAMETER_MIN_PROPERTY_COVERAGE

See Also:

Constant Field Values

PARAMETER_PROPERTY_LEARNING_RATE

public static final String PARAMETER_PROPERTY_LEARNING_RATE

See Also:

Constant Field Values

PARAMETER_OVERALL_PENALTY_WEIGHT

public static final String PARAMETER_OVERALL_PENALTY_WEIGHT

See Also:

Constant Field Values

PARAMETER_CHILDREN_PENALTY_WEIGHT

public static final String PARAMETER_CHILDREN_PENALTY_WEIGHT

See Also:

Constant Field Values

PARAMETER_COMPLEXITY_PENALTY_WEIGHT

public static final String PARAMETER_COMPLEXITY_PENALTY_WEIGHT

See Also:

Constant Field Values

PARAMETER_VERBOSE

public static final String PARAMETER_VERBOSE

See Also:

Constant Field Values

PARAMETER_ATOMIC_MEASURES

public static final String PARAMETER_ATOMIC_MEASURES

See Also:

Constant Field Values

PARAMETER_SAVE_MAPPING

public static final String PARAMETER_SAVE_MAPPING

See Also:

Constant Field Values

sourceVariable

protected String sourceVariable

targetVariable

protected String targetVariable

sourceUris

public static List<String> sourceUris

targetUris

public static List<String> targetUris

sourcePropertiesCoverageMap

protected Map<String,Double> sourcePropertiesCoverageMap

targetPropertiesCoverageMap

protected Map<String,Double> targetPropertiesCoverageMap

pseudoFMeasure

protected PseudoFMeasure pseudoFMeasure

trainingData

protected AMapping trainingData

isUnsupervised

protected boolean isUnsupervised

wombatParameterNames

protected Set<String> wombatParameterNames

refinementTreeRoot

protected Tree<RefinementNode> refinementTreeRoot

sourceSample

protected ACache sourceSample

targetSample

protected ACache targetSample

Constructor Detail

AWombat

protected AWombat()

Method Detail

createNode

protected RefinementNode createNode(AMapping mapping,
                                    String metricExpr)

Create new RefinementNode using either real or pseudo-F-Measure

Parameters:

mapping - of the node

metricExpr - learning specifications

Returns:

new RefinementNode

createNode

protected RefinementNode createNode(String metricExpr)

Parameters:

metricExpr - learning specifications

Returns:

new RefinementNode

executeAtomicMeasure

public AMapping executeAtomicMeasure(String sourceProperty,
                                     String targetProperty,
                                     String measure,
                                     double threshold)

Parameters:

sourceProperty - URI

targetProperty - URI

measure - name

threshold - of the LS

Returns:

Mapping from source to target resources after applying the atomic mapper measure(sourceProperty, targetProperty)

fMeasure

protected double fMeasure(AMapping predictions)

calculate either a real or a pseudo-F-Measure

Parameters:

predictions - Mapping

Returns:

F-measure

getMapingOfMetricExpression

protected AMapping getMapingOfMetricExpression(String metricExpression)

Looks first for the input metricExpression in the already constructed tree, if found the corresponding mapping is returned. Otherwise, the SetConstraintsMapper is generate the mapping from the metricExpression.

Parameters:

metricExpression - learning specifications

Returns:

Mapping corresponding to the input metric expression

getMapingOfMetricFromTree

protected AMapping getMapingOfMetricFromTree(String metricExpression,
                                             Tree<RefinementNode> r)

Parameters:

metricExpression - learning specifications

r - refinement tree

Returns:

return mapping of the input metricExpression from the search tree

getPredictions

protected AMapping getPredictions(LinkSpecification ls,
                                  ACache sCache,
                                  ACache tCache)

get mapping from source cache to target cache using metricExpression

Parameters:

ls - learning specifications

sCache - source Cache

tCache - target Cache

Returns:

Mapping from sCache to tCache

getPseudoFMeasure

public PseudoFMeasure getPseudoFMeasure()

getTrainingData

public AMapping getTrainingData()

init

protected void init(List<LearningParameter> lp,
                    ACache sourceCache,
                    ACache targetCache)

Description copied from class: ACoreMLAlgorithm

Initialize the core ML algorithm.

Overrides:

init in class ACoreMLAlgorithm

Parameters:

lp - learning parameters

sourceCache - the source cache

targetCache - the target cache

isUnsupervised

public boolean isUnsupervised()

precision

protected double precision(AMapping predictions)

calculate either a real or a pseudo-Precision

Parameters:

predictions - Mapping

Returns:

precision

recall

protected double recall(AMapping predictions)

calculate either a real or a pseudo-Recall

Parameters:

predictions - Mapping

Returns:

recall

findInitialClassifier

protected ExtendedClassifier findInitialClassifier(String sourceProperty,
                                                   String targetProperty,
                                                   String measure)

Computes the atomic classifiers by finding the highest possible F-measure achievable on a given property pair

Parameters:

sourceProperty - Property of source to use

targetProperty - Property of target to use

measure - Measure to be used

Returns:

Best simple classifier

setDefaultParameters

public void setDefaultParameters()

Description copied from class: ACoreMLAlgorithm

Set default ACoreMLAlgorithm parameters values

Specified by:

setDefaultParameters in class ACoreMLAlgorithm

isVerbose

protected boolean isVerbose()

getOverAllPenaltyWeight

protected double getOverAllPenaltyWeight()

getChildrenPenaltyWeight

protected double getChildrenPenaltyWeight()

getComplexityPenaltyWeight

protected double getComplexityPenaltyWeight()

getMinPropertyCoverage

protected double getMinPropertyCoverage()

getPropertyLearningRate

protected double getPropertyLearningRate()

getIterationTimeInMinutes

protected int getIterationTimeInMinutes()

getExcutionTimeInMinutes

protected int getExcutionTimeInMinutes()

getMaxFitnessThreshold

protected double getMaxFitnessThreshold()

getMaxIterationNumber

protected int getMaxIterationNumber()

getMaxRefinmentTreeSize

protected int getMaxRefinmentTreeSize()

getAtomicMeasures

protected Set<String> getAtomicMeasures()

fillSampleSourceTargetCaches

protected void fillSampleSourceTargetCaches(AMapping sample)

getMostPromisingNode

protected Tree<RefinementNode> getMostPromisingNode(Tree<RefinementNode> r,
                                                    double penaltyWeight)

Get the most promising node as the node with the best F-score

Parameters:

r - The whole refinement tree

penaltyWeight - penalty weight

Returns:

most promising node from the input tree r

findInitialClassifiers

protected List<ExtendedClassifier> findInitialClassifiers()

Returns:

initial classifiers

computePenalty

protected double computePenalty(Tree<RefinementNode> promesyChild)

Parameters:

promesyChild - promesy child

Returns:

children penalty + complexity penalty