Package se.hb.jcp.cp.measures

Class ObservedExcessCriterion

java.lang.Object

se.hb.jcp.cp.measures.AbstractSignificanceBasedMeasure

se.hb.jcp.cp.measures.ObservedExcessCriterion

All Implemented Interfaces:

IMeasure, IObservedMeasure

public class ObservedExcessCriterion
extends AbstractSignificanceBasedMeasure
implements IObservedMeasure

The OE/Observed Excess criterion is an observed efficiency measure based on the number of false labels in the label set. See [V. Vovk, V. Fedorova, I. Nouretdinov and A. Gammerman, "Criteria of Efficiency for Conformal Prediction", COPA 2016, LNAI 9653, pp. 23-39, 2016] for the definitions used here.

Author:

anders.gidenstam(at)hb.se

Constructor Summary

Constructors

Constructor	Description
ObservedExcessCriterion(double significanceLevel)	Creates an OE/Observed Excess criterion measure.

Method Summary

Modifier and Type	Method	Description
double	compute(ConformalClassification prediction, double trueLabel)	Computes the OE/Observed Excess criterion measure for this prediction and true label.

Methods inherited from class se.hb.jcp.cp.measures.AbstractSignificanceBasedMeasure

getName

Methods inherited from class java.lang.Object

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

Methods inherited from interface se.hb.jcp.cp.measures.IMeasure

getName

Constructor Detail

ObservedExcessCriterion

public ObservedExcessCriterion(double significanceLevel)

Creates an OE/Observed Excess criterion measure.

Parameters:

significanceLevel - the significance level used for the label sets.

Method Detail

compute

public double compute(ConformalClassification prediction,
                      double trueLabel)

Computes the OE/Observed Excess criterion measure for this prediction and true label.

Specified by:

compute in interface IObservedMeasure

Parameters:

prediction - the prediction.

trueLabel - the true label of the instance.

Returns:

the Observed Excess criterion measure for this prediction. Small values are preferable.