Package ai.libs.jaicore.ml.evaluation.evaluators.weka

Class ExtrapolatedSaturationPointEvaluator<I extends ILabeledAttributeArrayInstance<?>,D extends IOrderedLabeledAttributeArrayDataset<I,?>>

java.lang.Object

ai.libs.jaicore.ml.evaluation.evaluators.weka.ExtrapolatedSaturationPointEvaluator<I,D>

All Implemented Interfaces:

ai.libs.jaicore.basic.IObjectEvaluator<weka.classifiers.Classifier,java.lang.Double>, IClassifierEvaluator

public class ExtrapolatedSaturationPointEvaluator<I extends ILabeledAttributeArrayInstance<?>,D extends IOrderedLabeledAttributeArrayDataset<I,?>>
extends java.lang.Object
implements IClassifierEvaluator

For the classifier a learning curve will be extrapolated with a given set of anchorpoints. This learning curve can predict a saturation point with a tolerance epsilon. When a subsample is drawn at this saturation point it is the optimal trade-off between a fast training (therefore fast classifier evaluation) and dataset representability (therefore evaluation result expressiveness).

Constructor Summary

Constructors

Constructor	Description
ExtrapolatedSaturationPointEvaluator(int[] anchorpoints, ISamplingAlgorithmFactory<I,D,? extends ASamplingAlgorithm<I,D>> samplingAlgorithmFactory, D train, double trainSplitForAnchorpointsMeasurement, LearningCurveExtrapolationMethod extrapolationMethod, long seed, D test)	Create a classifier evaluator with an accuracy measurement at the extrapolated learning curves saturation point.

Method Summary

Modifier and Type	Method	Description
java.lang.Double	evaluate(weka.classifiers.Classifier classifier)
void	setEpsilon(double epsilon)

Methods inherited from class java.lang.Object

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

Constructor Detail

ExtrapolatedSaturationPointEvaluator

public ExtrapolatedSaturationPointEvaluator(int[] anchorpoints,
                                            ISamplingAlgorithmFactory<I,D,? extends ASamplingAlgorithm<I,D>> samplingAlgorithmFactory,
                                            D train,
                                            double trainSplitForAnchorpointsMeasurement,
                                            LearningCurveExtrapolationMethod extrapolationMethod,
                                            long seed,
                                            D test)

Create a classifier evaluator with an accuracy measurement at the extrapolated learning curves saturation point.

Parameters:

anchorpoints - Anchorpoints for the learning curve extrapolation.

samplingAlgorithmFactory - Subsampling factory for a subsampler to create samples at the given anchorpoints.

train - Dataset predict the learning curve with and where the subsample for the measurement is drawn from.

trainSplitForAnchorpointsMeasurement - Ratio to split the subsamples at the anchorpoints into train and test.

extrapolationMethod - Method to extrapolate a learning curve from the accuracy measurements at the anchorpoints.

seed - Random seed.

test - Test dataset to measure the accuracy.

Method Detail

setEpsilon

public void setEpsilon(double epsilon)

evaluate

public java.lang.Double evaluate(weka.classifiers.Classifier classifier)
                          throws java.lang.InterruptedException,
                                 ai.libs.jaicore.basic.algorithm.exceptions.ObjectEvaluationFailedException

Specified by:

evaluate in interface ai.libs.jaicore.basic.IObjectEvaluator<I extends ILabeledAttributeArrayInstance<?>,D extends IOrderedLabeledAttributeArrayDataset<I,?>>

Throws:

java.lang.InterruptedException

ai.libs.jaicore.basic.algorithm.exceptions.ObjectEvaluationFailedException