org.deeplearning4j.eval.curves

Class PrecisionRecallCurve

java.lang.Object

org.deeplearning4j.eval.curves.BaseCurve

org.deeplearning4j.eval.curves.PrecisionRecallCurve

public class PrecisionRecallCurve
extends BaseCurve

Precision recall curve: A set of (recall, precision) points and different thresholds

Author:

Alex Black

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	PrecisionRecallCurve.Confusion
static class	PrecisionRecallCurve.Point

Field Summary

Fields inherited from class org.deeplearning4j.eval.curves.BaseCurve

DEFAULT_FORMAT_PREC

Constructor Summary

Constructors

Constructor and Description
PrecisionRecallCurve(double[] threshold, double[] precision, double[] recall, int[] tpCount, int[] fpCount, int[] fnCount, int totalCount)

Method Summary

Modifier and Type	Method and Description
double	calculateAUPRC()
static PrecisionRecallCurve	fromJson(String json)
static PrecisionRecallCurve	fromYaml(String yaml)
PrecisionRecallCurve.Confusion	getConfusionMatrixAtPoint(int point) Get the binary confusion matrix for the given position.
PrecisionRecallCurve.Confusion	getConfusionMatrixAtThreshold(double threshold) Get the binary confusion matrix for the given threshold.
PrecisionRecallCurve.Point	getPointAtPrecision(double precision) Get the point (index, threshold, precision, recall) at the given precision. Specifically, return the points at the lowest threshold that has precision equal to or greater than the requested precision.
PrecisionRecallCurve.Point	getPointAtRecall(double recall) Get the point (index, threshold, precision, recall) at the given recall. Specifically, return the points at the highest threshold that has recall equal to or greater than the requested recall.
PrecisionRecallCurve.Point	getPointAtThreshold(double threshold) Get the point (index, threshold, precision, recall) at the given threshold. Note that if the threshold is not found exactly, the next highest threshold exceeding the requested threshold is returned
double	getPrecision(int i)
double	getRecall(int i)
double	getThreshold(int i)
String	getTitle()
double[]	getX()
double[]	getY()
int	numPoints()

Methods inherited from class org.deeplearning4j.eval.curves.BaseCurve

calculateArea, calculateArea, format, fromJson, fromYaml, toJson, toYaml

Methods inherited from class java.lang.Object

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

Constructor Detail

PrecisionRecallCurve

public PrecisionRecallCurve(double[] threshold,
                            double[] precision,
                            double[] recall,
                            int[] tpCount,
                            int[] fpCount,
                            int[] fnCount,
                            int totalCount)

Method Detail

numPoints

public int numPoints()

Specified by:

numPoints in class BaseCurve

Returns:

The number of points in the curve

getX

public double[] getX()

Specified by:

getX in class BaseCurve

Returns:

X axis values

getY

public double[] getY()

Specified by:

getY in class BaseCurve

Returns:

Y-axis values

getTitle

public String getTitle()

Specified by:

getTitle in class BaseCurve

Returns:

Title for the curve

getThreshold

public double getThreshold(int i)

Parameters:

i - Point number, 0 to numPoints()-1 inclusive

Returns:

Threshold of a given point

getPrecision

public double getPrecision(int i)

Parameters:

i - Point number, 0 to numPoints()-1 inclusive

Returns:

Precision of a given point

getRecall

public double getRecall(int i)

Parameters:

i - Point number, 0 to numPoints()-1 inclusive

Returns:

Recall of a given point

calculateAUPRC

public double calculateAUPRC()

Returns:

The area under the precision recall curve

getPointAtThreshold

public PrecisionRecallCurve.Point getPointAtThreshold(double threshold)

Get the point (index, threshold, precision, recall) at the given threshold.
Note that if the threshold is not found exactly, the next highest threshold exceeding the requested threshold is returned

Parameters:

threshold - Threshold to get the point for

Returns:

point (index, threshold, precision, recall) at the given threshold

getPointAtPrecision

public PrecisionRecallCurve.Point getPointAtPrecision(double precision)

Get the point (index, threshold, precision, recall) at the given precision.
Specifically, return the points at the lowest threshold that has precision equal to or greater than the requested precision.

Parameters:

precision - Precision to get the point for

Returns:

point (index, threshold, precision, recall) at (or closest exceeding) the given precision

getPointAtRecall

public PrecisionRecallCurve.Point getPointAtRecall(double recall)

Get the point (index, threshold, precision, recall) at the given recall.
Specifically, return the points at the highest threshold that has recall equal to or greater than the requested recall.

Parameters:

recall - Recall to get the point for

Returns:

point (index, threshold, precision, recall) at (or closest exceeding) the given recall

getConfusionMatrixAtThreshold

public PrecisionRecallCurve.Confusion getConfusionMatrixAtThreshold(double threshold)

Get the binary confusion matrix for the given threshold. As per getPointAtThreshold(double), if the threshold is not found exactly, the next highest threshold exceeding the requested threshold is returned

Parameters:

threshold - Threshold at which to get the confusion matrix

Returns:

Binary confusion matrix

getConfusionMatrixAtPoint

public PrecisionRecallCurve.Confusion getConfusionMatrixAtPoint(int point)

Get the binary confusion matrix for the given position. As per getPointAtThreshold(double).

Parameters:

point - Position at which to get the binary confusion matrix

Returns:

Binary confusion matrix

fromJson

public static PrecisionRecallCurve fromJson(String json)

fromYaml

public static PrecisionRecallCurve fromYaml(String yaml)