Skip to contents

Customizing plots

Source: vignettes/articles/plots-customizing.Rmd
plots-customizing.Rmd

autoplot() returns an ordinary ggplot object, so the whole of ggplot2 applies to it.

library(precrec)
library(ggplot2)

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

Add layers to the result 

autoplot(curves, "PRC") +
  labs(title = "Precision-recall", subtitle = "P10N10") +
  theme_minimal()

Draw the baseline yourself

The plots leave the precision-recall baseline out, because it sits at the proportion of positives and that differs between datasets. When you are looking at one dataset you know the number, so add it.

baseline <- sum(P10N10$labels == 1) / length(P10N10$labels)

autoplot(curves, "PRC") +
  geom_hline(yintercept = baseline, linetype = "dashed", color = "grey40")

Start from the data instead

fortify() is the ggplot2 hook, so a precrec object can go straight into ggplot() when you want to build the plot from scratch.

df <- fortify(curves)
head(df)
#>       x   y modname dsid dsid_modname curvetype
#> 1 0.000 0.0      m1    1         m1:1       ROC
#> 2 0.000 0.1      m1    1         m1:1       ROC
#> 3 0.000 0.2      m1    1         m1:1       ROC
#> 4 0.001 0.2      m1    1         m1:1       ROC
#> 5 0.002 0.2      m1    1         m1:1       ROC
#> 6 0.003 0.2      m1    1         m1:1       ROC

ggplot(subset(df, curvetype == "PRC"), aes(x = x, y = y)) +
  geom_line(color = "steelblue", linewidth = 1) +
  coord_fixed() +
  labs(x = "Recall", y = "Precision") +
  theme_bw()

The columns are x, y, modname, dsid, dsid_modname and curvetype - already in long form, so mapping color or facets to a model is direct.

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

ggplot(subset(mdf, curvetype == "PRC"), aes(x = x, y = y, color = modname)) +
  geom_line() +
  facet_wrap(~modname) +
  theme_bw() +
  theme(legend.position = "none")

Getting the grob

Multi-panel output - both curve types at once, or several measures - is assembled from separate plots. ret_grob = TRUE returns that assembled object instead of drawing it, for placing in a larger layout.

g <- autoplot(curves, ret_grob = TRUE)

Base graphics

plot() takes the usual base arguments, so titles, colors and line types work the way they do everywhere else.

plot(curves, "PRC")

Next