Skip to contents

Partial curves

Source: vignettes/articles/plots-partial-curves.Rmd
plots-partial-curves.Rmd

Often only part of a curve matters - the low-false-positive end when every alert costs a review, or the high-recall end when a miss is expensive. part() restricts a curve to that range.

library(precrec)
library(ggplot2)

curves <- evalmod(scores = P10N10$scores, labels = P10N10$labels)

Restricting the range 

partial <- part(curves, xlim = c(0, 0.25))

autoplot(partial)

The full curve stays visible in outline and the selected part is highlighted, so the region is read in context rather than in isolation.

ylim restricts the other axis, and both can be given at once.

partial_y <- part(curves, ylim = c(0.5, 1))

autoplot(partial_y)

The areas

pauc() returns the area over the restricted range.

knitr::kable(pauc(partial))
modnames dsids curvetypes paucs spaucs
m1 1 ROC 0.1006250 0.4025000
m1 1 PRC 0.2345849 0.9383396

paucs is the raw area, which is small simply because the range is narrow. spaucs is the standardized version, rescaled to 0 to 1 so that ranges of different widths can be compared with each other and against a full AUC.

Report the standardized one unless you have a reason not to. A partial AUC of 0.18 sounds poor and may be excellent for the quarter of the axis it covers.

Several models

part() works on any evalmod() object, so the comparison carries over.

samps <- create_sim_samples(1, 100, 100, "all")
mdat <- mmdata(samps[["scores"]], samps[["labels"]],
  modnames = samps[["modnames"]]
)

mpartial <- part(evalmod(mdat), xlim = c(0, 0.25))

knitr::kable(pauc(mpartial))
modnames dsids curvetypes paucs spaucs
random 1 ROC 0.0277000 0.1108000
random 1 PRC 0.1257566 0.5030262
poor_er 1 ROC 0.1208000 0.4832000
poor_er 1 PRC 0.2091682 0.8366726
good_er 1 ROC 0.1511000 0.6044000
good_er 1 PRC 0.2500000 1.0000000
excel 1 ROC 0.2284000 0.9136000
excel 1 PRC 0.2500000 1.0000000
perf 1 ROC 0.2500000 1.0000000
perf 1 PRC 0.2500000 1.0000000

Next