de.jstacs.classifiers.performanceMeasures
Class AbstractPerformanceMeasure

java.lang.Object
  de.jstacs.parameters.ParameterSet
      de.jstacs.classifiers.performanceMeasures.AbstractPerformanceMeasure

All Implemented Interfaces:

PerformanceMeasure, GalaxyConvertible, Storable, Cloneable

Direct Known Subclasses:

AbstractTwoClassPerformanceMeasure, ClassificationRate, ConfusionMatrix

public abstract class AbstractPerformanceMeasure
extends ParameterSet
implements PerformanceMeasure

This class is the abstract super class of any performance measure used to evaluate an AbstractClassifier. It is recommended to use the method AbstractClassifier.evaluate(AbstractPerformanceMeasureParameterSet, boolean, de.jstacs.data.DataSet...) for evaluating the performance of any classifier.

Author:

Jan Grau, Jens Keilwagen

See Also:

AbstractClassifier, PerformanceMeasureParameterSet

Nested Class Summary

Nested classes/interfaces inherited from class de.jstacs.parameters.ParameterSet

ParameterSet.ParameterList

Field Summary

Fields inherited from class de.jstacs.parameters.ParameterSet

errorMessage, parameters, parent

Constructor Summary

protected	AbstractPerformanceMeasure() Constructs a new AbstractPerformanceMeasure with empty parameter values.
protected	AbstractPerformanceMeasure(StringBuffer xml) The standard constructor for the interface Storable.

Method Summary

ResultSet	compute(double[][][] classSpecificScores) This method allows to compute the performance measure of given class specific scores.
ResultSet	compute(double[] sortedScoresClass0, double[] sortedScoresClass1) This method allows to compute the performance measure of given sorted score ratios.
protected static int	findSplitIndex(double[] sortedScores, double t) Returns the index in sortedScores with value greater or equal to t.
protected static double	findThreshold(double[] sortedReferenceScores, double[] sortedMeasureScores, double[] referenceWeights, double percentage, boolean atLeast) Determines the threshold for a given percentage on the reference weights using the scores in sortedReferenceScores and sortedMeasureScores.
static SelectionParameter	getCollectionOfAllMeasures(int numClasses, boolean numerical) This method creates an instance of an SelectionParameter that can be used to create an instance of PerformanceMeasureParameterSet or NumericalPerformanceMeasureParameterSet.
abstract String	getName() The method returns the name of the performance measure.
protected static double	getWeight(double[] weight, int index) Returns the weight at index in weight or 1 if weight is null.
static boolean	simpleWeights(double[] weight) Returns true if all weights in weight are 1.

Methods inherited from class de.jstacs.parameters.ParameterSet

clone, fromGalaxy, fromXML, getAllParameterNames, getComment, getComment, getErrorMessage, getIndex, getName, getName, getNumberOfParameters, getParameterAt, getParameterForName, getParent, hasDefaultOrIsSet, initParameterList, initParameterList, isAtomic, isComparable, parametersLoaded, reset, setParent, toGalaxy, toXML

Methods inherited from class java.lang.Object

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

Methods inherited from interface de.jstacs.classifiers.performanceMeasures.PerformanceMeasure

compute, compute, getAllowedNumberOfClasses

Constructor Detail

AbstractPerformanceMeasure

protected AbstractPerformanceMeasure()

Constructs a new AbstractPerformanceMeasure with empty parameter values.

AbstractPerformanceMeasure

protected AbstractPerformanceMeasure(StringBuffer xml)
                              throws NonParsableException

The standard constructor for the interface Storable. Constructs a AbstractPerformanceMeasure out of an XML representation.

Parameters:

xml - the XML representation as StringBuffer

Throws:

NonParsableException - if the AbstractPerformanceMeasure could not be reconstructed out of the StringBuffer xml

Method Detail

getName

public abstract String getName()

Description copied from interface: PerformanceMeasure

The method returns the name of the performance measure.

Specified by:

getName in interface PerformanceMeasure

Returns:

the name of the performance measure

compute

public ResultSet compute(double[] sortedScoresClass0,
                         double[] sortedScoresClass1)

Description copied from interface: PerformanceMeasure

This method allows to compute the performance measure of given sorted score ratios. This method can only be used for binary classifiers.

Specified by:

compute in interface PerformanceMeasure

Parameters:

sortedScoresClass0 - the sorted score ratios of class 0

sortedScoresClass1 - the sorted score ratios of class 1

Returns:

a result set containing the results of the performance measure

See Also:

PerformanceMeasure.compute(double[], double[], double[], double[])

compute

public ResultSet compute(double[][][] classSpecificScores)

Description copied from interface: PerformanceMeasure

This method allows to compute the performance measure of given class specific scores.

Specified by:

compute in interface PerformanceMeasure

Parameters:

classSpecificScores - the scores; first dimension = data sets, second dimension = sequences of the data set, third dimension classes of the classifier

Returns:

a result set containing the results of the performance measure

See Also:

PerformanceMeasure.compute(double[][][], double[][])

getCollectionOfAllMeasures

public static SelectionParameter getCollectionOfAllMeasures(int numClasses,
                                                            boolean numerical)
                                                     throws Exception

This method creates an instance of an SelectionParameter that can be used to create an instance of PerformanceMeasureParameterSet or NumericalPerformanceMeasureParameterSet.

Parameters:

numClasses - the number of classes

numerical - a switch indicating whether all performance measures or only those implementing NumericalPerformanceMeasure shall be contained in the returned SelectionParameter

Returns:

a SelectionParameter that can be used to create an instance of PerformanceMeasureParameterSet or NumericalPerformanceMeasureParameterSet

Throws:

Exception - if something went wrong, e.g. missing empty constructor of any performance measure.

See Also:

AbstractClassifier.getNumberOfClasses()

findThreshold

protected static double findThreshold(double[] sortedReferenceScores,
                                      double[] sortedMeasureScores,
                                      double[] referenceWeights,
                                      double percentage,
                                      boolean atLeast)

Determines the threshold for a given percentage on the reference weights using the scores in sortedReferenceScores and sortedMeasureScores.

Parameters:

sortedReferenceScores - the scores of the reference

sortedMeasureScores - the scores to be thresholded

referenceWeights - the weights on the entries of the reference scores

percentage - the percentage

atLeast - if we do not meet the percentage exactly, shall the actual percentage be at least the given one

Returns:

the threshold

findSplitIndex

protected static int findSplitIndex(double[] sortedScores,
                                    double t)

Returns the index in sortedScores with value greater or equal to t.

Parameters:

sortedScores - the array of sorted values

t - the threshold

Returns:

the index in sortedScores with value greater or equal to t

getWeight

protected static final double getWeight(double[] weight,
                                        int index)

Returns the weight at index in weight or 1 if weight is null.

Parameters:

weight - the weights

index - the index

Returns:

the weight at index or 1

simpleWeights

public static boolean simpleWeights(double[] weight)

Returns true if all weights in weight are 1.

Parameters:

weight - the weights

Returns:

if all weights in weight are 1