mulan.evaluation.measure
Interface Measure

All Known Implementing Classes:

AveragePrecision, BipartitionMeasureBase, ConfidenceMeasureBase, Coverage, ErrorSetSize, ExampleBasedAccuracy, ExampleBasedBipartitionMeasureBase, ExampleBasedFMeasure, ExampleBasedPrecision, ExampleBasedRecall, ExampleBasedSpecificity, GeometricMeanAverageInterpolatedPrecision, GeometricMeanAveragePrecision, HammingLoss, HierarchicalLoss, IsError, LabelBasedAUC, LabelBasedAveragePrecision, LabelBasedBipartitionMeasureBase, LabelBasedFMeasure, LabelBasedPrecision, LabelBasedRecall, LabelBasedSpecificity, LossBasedBipartitionMeasureBase, LossBasedRankingMeasureBase, MacroAUC, MacroFMeasure, MacroPrecision, MacroRecall, MacroSpecificity, MeanAverageInterpolatedPrecision, MeanAveragePrecision, MeasureBase, MicroAUC, MicroFMeasure, MicroPrecision, MicroRecall, MicroSpecificity, OneError, RankingLoss, RankingMeasureBase, SubsetAccuracy

public interface Measure

Interface for a measure, used to evaluate the performance of a multi-label learner on when performing multi-label learning task. Various measures capture different characteristics of a learning task performance.

Author:

Jozef Vilcek

Method Summary

double	getIdealValue() Gets an 'ideal' value of a measure.
String	getName() Gets the name of a measure.
double	getValue() Gets the value of a measure.
Measure	makeCopy() Creates a deep copy of the given measure using serialization.
void	reset() Resets the cumulated measure value, so the process of computation can be started from beginning (e.g. for a new series of outputs from learning task).
void	update(MultiLabelOutput prediction, boolean[] truth) Computes the value of a measure for the given prediction and true labels.

Method Detail

getName

String getName()

Gets the name of a measure.

Returns:

the name of a measure.

getValue

double getValue()

Gets the value of a measure. The measure is incrementally cumulated for learner's prediction by each update(MultiLabelOutput, boolean[]) call. The value returned by the method, returns sum of all update calls divided by the number of calls (average of all measures for all predictions).

Returns:

the average measure value computed so far

getIdealValue

double getIdealValue()

Gets an 'ideal' value of a measure. The 'ideal' means, that the value represents the best achievable performance of a learner for an prediction of a multi-label task and associated true labels.

Returns:

the ideal value

update

void update(MultiLabelOutput prediction,
            boolean[] truth)

Computes the value of a measure for the given prediction and true labels. The immediate value of a measure is returned and result is added to the cumulated measure value.

Parameters:

prediction - the prediction for which measure has to be computed

truth - the true labels bipartition for given prediction

See Also:

getValue()

makeCopy

Measure makeCopy()
                 throws Exception

Creates a deep copy of the given measure using serialization.

Returns:

a deep copy of the measure

Throws:

Exception - if an error occurs while making copy of the measure.

reset

void reset()

Resets the cumulated measure value, so the process of computation can be started from beginning (e.g. for a new series of outputs from learning task).