org.nd4j.evaluation

Class EvaluationUtils

java.lang.Object

org.nd4j.evaluation.EvaluationUtils

public class EvaluationUtils
extends Object

Constructor Summary

Constructors

Constructor and Description
EvaluationUtils()

Method Summary

Modifier and Type	Method and Description
static Pair<INDArray,INDArray>	extractNonMaskedTimeSteps(INDArray labels, INDArray predicted, INDArray outputMask)
static double	falseNegativeRate(long fnCount, long tpCount, double edgeCase) Calculate the false negative rate from the false negative counts and true positive count
static double	falsePositiveRate(long fpCount, long tnCount, double edgeCase) Calculate the false positive rate from the false positive count and true negative count
static double	fBeta(double beta, double precision, double recall) Calculate the F-beta value from precision and recall
static double	fBeta(double beta, long tp, long fp, long fn) Calculate the F beta value from counts
static double	gMeasure(double precision, double recall) Calculate the G-measure from precision and recall
static double	matthewsCorrelation(long tp, long fp, long fn, long tn) Calculate the binary Matthews correlation coefficient from counts
static double	precision(long tpCount, long fpCount, double edgeCase) Calculate the precision from true positive and false positive counts
static double	recall(long tpCount, long fnCount, double edgeCase) Calculate the recall from true positive and false negative counts
static INDArray	reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask) Reshape time series mask arrays.
static INDArray	reshapeTimeSeriesTo2d(INDArray labels)

Methods inherited from class java.lang.Object

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

Constructor Detail

EvaluationUtils

public EvaluationUtils()

Method Detail

precision

public static double precision(long tpCount,
                               long fpCount,
                               double edgeCase)

Calculate the precision from true positive and false positive counts

Parameters:

tpCount - True positive count

fpCount - False positive count

edgeCase - Edge case value use to avoid 0/0

Returns:

Precision

recall

public static double recall(long tpCount,
                            long fnCount,
                            double edgeCase)

Calculate the recall from true positive and false negative counts

Parameters:

tpCount - True positive count

fnCount - False negative count

edgeCase - Edge case values used to avoid 0/0

Returns:

Recall

falsePositiveRate

public static double falsePositiveRate(long fpCount,
                                       long tnCount,
                                       double edgeCase)

Calculate the false positive rate from the false positive count and true negative count

Parameters:

fpCount - False positive count

tnCount - True negative count

edgeCase - Edge case values are used to avoid 0/0

Returns:

False positive rate

falseNegativeRate

public static double falseNegativeRate(long fnCount,
                                       long tpCount,
                                       double edgeCase)

Calculate the false negative rate from the false negative counts and true positive count

Parameters:

fnCount - False negative count

tpCount - True positive count

edgeCase - Edge case value to use to avoid 0/0

Returns:

False negative rate

fBeta

public static double fBeta(double beta,
                           long tp,
                           long fp,
                           long fn)

Calculate the F beta value from counts

Parameters:

beta - Beta of value to use

tp - True positive count

fp - False positive count

fn - False negative count

Returns:

F beta

fBeta

public static double fBeta(double beta,
                           double precision,
                           double recall)

Calculate the F-beta value from precision and recall

Parameters:

beta - Beta value to use

precision - Precision

recall - Recall

Returns:

F-beta value

gMeasure

public static double gMeasure(double precision,
                              double recall)

Calculate the G-measure from precision and recall

Parameters:

precision - Precision value

recall - Recall value

Returns:

G-measure

matthewsCorrelation

public static double matthewsCorrelation(long tp,
                                         long fp,
                                         long fn,
                                         long tn)

Calculate the binary Matthews correlation coefficient from counts

Parameters:

tp - True positive count

fp - False positive counts

fn - False negative counts

tn - True negative count

Returns:

Matthews correlation coefficient

reshapeTimeSeriesTo2d

public static INDArray reshapeTimeSeriesTo2d(INDArray labels)

extractNonMaskedTimeSteps

public static Pair<INDArray,INDArray> extractNonMaskedTimeSteps(INDArray labels,
                                                                INDArray predicted,
                                                                INDArray outputMask)

reshapeTimeSeriesMaskToVector

public static INDArray reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask)

Reshape time series mask arrays. This should match the assumptions (f order, etc) in RnnOutputLayer

Parameters:

timeSeriesMask - Mask array to reshape to a column vector

Returns:

Mask array as a column vector