org.tribuo.classification.evaluation

Class LabelEvaluationUtil

java.lang.Object

org.tribuo.classification.evaluation.LabelEvaluationUtil

public final class LabelEvaluationUtil
extends Object

Static utility functions for calculating performance metrics on Labels.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LabelEvaluationUtil.PRCurve Stores the Precision-Recall curve as three arrays: the precisions, the recalls, and the thresholds associated with those values.
static class	LabelEvaluationUtil.ROC Stores the ROC curve as three arrays: the false positive rate, the true positive rate, and the thresholds associated with those rates.

Method Summary

Modifier and Type	Method and Description
static double	averagedPrecision(boolean[] yPos, double[] yScore) Summarises a Precision-Recall Curve by taking the weighted mean of the precisions at a given threshold, where the weight is the recall achieved at that threshold.
static double	binaryAUCROC(boolean[] yPos, double[] yScore) Calculates the area under the receiver operator characteristic curve, i.e., the AUC of the ROC curve.
static LabelEvaluationUtil.PRCurve	generatePRCurve(boolean[] yPos, double[] yScore) Calculates the Precision Recall curve for a single label.
static LabelEvaluationUtil.ROC	generateROCCurve(boolean[] yPos, double[] yScore) Calculates the binary ROC for a single label.

Methods inherited from class java.lang.Object

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

Method Detail

averagedPrecision

public static double averagedPrecision(boolean[] yPos,
                                       double[] yScore)

Summarises a Precision-Recall Curve by taking the weighted mean of the precisions at a given threshold, where the weight is the recall achieved at that threshold. Follows scikit-learn's implementation. In general use the AUC for a Precision-Recall Gain curve as the area under the precision-recall curve is not properly normalized.

Parameters:

yPos - Each element is true if the label was from the positive class.

yScore - Each element is the score of the positive class.

Returns:

The averaged precision.

generatePRCurve

public static LabelEvaluationUtil.PRCurve generatePRCurve(boolean[] yPos,
                                                          double[] yScore)

Calculates the Precision Recall curve for a single label. In general use Precision-Recall Gain curves.

Parameters:

yPos - Each element is true if the label was from the positive class.

yScore - Each element is the score of the positive class.

Returns:

The PRCurve for one label.

binaryAUCROC

public static double binaryAUCROC(boolean[] yPos,
                                  double[] yScore)

Calculates the area under the receiver operator characteristic curve, i.e., the AUC of the ROC curve.

Parameters:

yPos - Is the associated index a positive label.

yScore - The score of the positive class.

Returns:

The auc (a value bounded 0.0-1.0).

generateROCCurve

public static LabelEvaluationUtil.ROC generateROCCurve(boolean[] yPos,
                                                       double[] yScore)

Calculates the binary ROC for a single label.

Parameters:

yPos - Each element is true if the label was from the positive class.

yScore - Each element is the score of the positive class.

Returns:

The ROC for one label.