moa.core.utils

Class EvalUtils

java.lang.Object

moa.core.utils.EvalUtils

public abstract class EvalUtils
extends Object

Evaluation Utilities.

Version:

$Revision: 1 $

Author:

Jesse Read (jesse@tsc.uc3m.es)

Constructor Summary

Constructors

Constructor and Description
EvalUtils()

Method Summary

Methods

Modifier and Type	Method and Description
static double	calcLogLoss(double R, double P, double C) Calculate Log Loss.
static double	calibrateThreshold(ArrayList<double[]> Y, double LC_train) Calibrate a threshold.
static double[]	calibrateThresholds(ArrayList<double[]> Y, double[] LC_train) Calibrate thresholds.
static HashMap<String,Double>	evaluateMultiLabel(ArrayList<double[]> Distributions, ArrayList<int[]> TrueRelevances, double t) Calculate Performance Measures.
static int[]	toIntArray(weka.core.Instance x, int L) Convert Instance to bit array.

Methods inherited from class java.lang.Object

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

Constructor Detail

EvalUtils

public EvalUtils()

Method Detail

toIntArray

public static final int[] toIntArray(weka.core.Instance x,
               int L)

Convert Instance to bit array. Returns the labels of this instance as an int array relevance representation.

calibrateThreshold

public static double calibrateThreshold(ArrayList<double[]> Y,
                        double LC_train)

Calibrate a threshold. Based upon a supplied label cardinality from the training set.

calibrateThresholds

public static double[] calibrateThresholds(ArrayList<double[]> Y,
                           double[] LC_train)

Calibrate thresholds. One threshold for each label, based on training-set label cardinality. Assumes that each double[] in Y is of length LC_train.length;

evaluateMultiLabel

public static HashMap<String,Double> evaluateMultiLabel(ArrayList<double[]> Distributions,
                                        ArrayList<int[]> TrueRelevances,
                                        double t)

Calculate Performance Measures. Given a threshold t on a batch of predicted Distributions and their corresponding TrueRelevances. (Rankings.size() must equal Actuals.size()).

calcLogLoss

public static double calcLogLoss(double R,
                 double P,
                 double C)

Calculate Log Loss. See Jesse Read. Scalable Multi-label Classification. PhD thesis, University of Waikato, 2010.