Skip to contents

Precision-recall curve

Source: R/prediction.R
plot_pr_curve.Rd

Precision against recall for a binary outcome. On an imbalanced outcome the ROC curve can look encouraging while precision stays low, which is what this plot exposes: it is the recommended companion when the event is rare.

Usage

plot_pr_curve(
  scores,
  labels,
  positive = NULL,
  palette = "house",
  title = NULL,
  font = "Arial",
  show_auc = TRUE
)

Arguments

scores

Numeric vector, or a matrix or data frame with one column per marker.

labels

Binary outcome, one value per observation.

positive

Percentile of the scores above which the positive class is expected. Used only to draw the no-skill reference line, which is the prevalence of the positive class.

palette

Palette name passed to get_color().

title

Plot title.

font

Font family used in the plot.

show_auc

Print the area under the curve, computed by the trapezoid rule, in the legend.

Value

A ggplot object.

See also

plot_ROC() for discrimination, plot_calibration() for the level of the predictions.

Examples

set.seed(1)
outcome <- rbinom(200, 1, 0.15)
score <- outcome * 0.8 + rnorm(200)
p <- plot_pr_curve(score, outcome)