Specify asymetrical breaks in `scale_fill_stepsn()`

I want to add a stepped gradient for the fill aesthetic, with specific colors for specific intervals.

For instance, I'd like red between 0 and 15, blue between 15 and 18, etc.

Here is the closest I've (painfully) come to my goal:

library(tidyverse)
val_breaks = c(0, 15, 18, 20, 25)
col_breaks = c("red", "blue", "yellow", "purple")

#4 intervals, 4 colors
cut(mtcars$qsec, breaks=val_breaks) %>% levels()
#> [1] "(0,15]"  "(15,18]" "(18,20]" "(20,25]"

mtcars %>%
  ggplot(aes(x = mpg, y = wt, color = qsec)) +
  geom_point(size = 3) +
  scale_color_stepsn(
    colors=col_breaks,
    limits=c(0,25),
    breaks=val_breaks,
    values=scales::rescale(val_breaks)
  )

^{Created on 2024-12-08 with reprex v2.1.1}

As you can see, the colors are not really respected and mix with each other.

Is there a way?

You can achieve your desired result by passing the midpoints of the intervals to the values= argument and by setting the from= argument of rescale according to the limits= used for the scale:

library(tidyverse)

val_breaks <- c(0, 15, 18, 20, 25)
col_breaks <- c("red", "blue", "yellow", "purple")

midpoints <- val_breaks[-1] - diff(val_breaks) / 2

mtcars %>%
  ggplot(aes(x = mpg, y = wt, color = qsec)) +
  geom_point(size = 3) +
  scale_color_stepsn(
    colors = col_breaks,
    limits = c(0, 25),
    breaks = val_breaks,
    values = scales::rescale(
      midpoints,
      from = c(0, 25)
    )
  )

Explanation

The reason is that a binned scale is still a continuous scale and the color palette is still a color gradient but the color assigned to all points in a bin is based on the midpoint of the interval or bin.

With scales::rescale(val_breaks) you specified that the colors apply to the edges of the bins. But as the color for each bin is based on the midpoint you end up with a mix of the colors at the bins edges, i.e.

pal <- scales::pal_gradient_n(
  col_breaks,
  scales::rescale(val_breaks),
  "Lab"
)

pal(scales::rescale(midpoints, from = range(val_breaks))) |>
  scales::show_col()

Hence, to fix the issue we have to make sure that the colors apply to the midpoints:

pal2 <- scales::pal_gradient_n(
  col_breaks,
  scales::rescale(midpoints, from = range(val_breaks)),
  "Lab"
)

pal2(scales::rescale(midpoints, from = range(val_breaks))) |>
  scales::show_col()