AUPRC: Area Under the Precision Recall Curve

In PerfMeas: Performance Measures for Ranking and Classification Tasks

Area Under the Precision Recall Curve

Description

Functions to compute the Area Under the Precision Recall Curve (AUPRC) and the Area Under the F-score Recall Curve (AUFRC)

Usage

AUPRC(z, comp.precision=TRUE)
trap.rule.integral(x,y)

Arguments

z	a list of lists. The components of the outer list is a list returned from the function precision.at.all.recall.levels that reports precision, recall and f-score results at different levels for different methods or tasks.
comp.precision	boolean. It TRUE (default) the AUPRC is computed otherwise the area under the F-score curve is computed
x	vector of the x values in increasing order
y	vector of the corresponding y=f(x) values

Details

AUPRC computes the Area Under the Precision Recall Curve or the Area Under the F-score Recall Curve (AUFRC) for multiple curves by using the output of the function precision.at.all.recall.levels.

The function trap.rule.integral implements the trapezoidal rule of integration and can be used to compute the integral of any empirical function expressed as a set of pair values (a vector of x values and a vector of y = f(x) values). In particular if x is the recall (with values in ascending order) and y the corresponding precision, trap.rule.integral copmutes the AUPRC.

Value

AUPRC returns the value of the AUPRC (if the argument comp.precision = TRUE), otherwise the value of the AUFRC.

trap.rule.integral returns the value of the integral.

See Also

AUC.measures, PXR

Examples

# loading matrices of scores an correponding table of classes
data(T);
data(Scores);
res=list();
classes=1:10
# computing precision recall values
for (j in classes) res=c(res, list(precision.at.all.recall.levels(Scores[,j], T[,j])));
names(res)<-seq(0.1, 1, by=0.1);
# computing AUPRC
AUPRC (res, comp.precision=TRUE);
# computing AU F-score recall curve
AUPRC (res, comp.precision=TRUE);

# Loading precision at given recall levels for different methods
data(PrecRec);
# computing AUPRC for different methods
x <- seq(0.1, 1, by=0.1);
res <- numeric(nrow(PrecRec));
names(res) <- rownames(PrecRec);
for (i in 1:nrow(PrecRec))  
  res[i] <- trap.rule.integral(x, PrecRec[i,]);
print(res);

PerfMeas documentation built on Sept. 14, 2022, 9:06 a.m.