stat_function disregards log and possibly other scale

[arwarw]

stat_function() seems to disregard scale_y_log10() and similar. There has been a PR #1011 that is (as far as I understand) supposed to have fixed this issue, merged in 2015 and released in 1.0.1.9000.

However, I can observe the following behavior in ggplot2 3.2.1 and 3.3.0 (did not test others):

library(tidyverse)
fn <- function(x) {exp(x)}
data <- tibble(x = c(1:10), y=exp(x))
data %>% ggplot(aes(x, y)) + geom_point() + stat_function(fun=fn)

This (above) looks as expected, nothing was rescaled, the function matches the data points.

data %>% ggplot(aes(x, y)) + geom_point() + stat_function(fun=fn) + scale_y_log10()

The problem can be observed here, while the data points were adapted to a logscale y axis, the function wasn't.

data %>% ggplot(aes(x, y)) + geom_point() + stat_function(fun=function(x){log10(fn(x))}) + scale_y_log10()

This is what I would the plot to look like (just without the workaround of applying the extra log10()).

^{Created on 2020-03-21 by the reprex package (v0.3.0)}

[clauswilke]

I vaguely recall that this also has bothered me in the past. Let's see if we can figure out a way to fix this.

[clauswilke]

Looking at the code, it seems to me that transformations along the x axis are handled correctly but transformations along y are not:

ggplot2/R/stat-function.r

Lines 109 to 111 in da2388e

	# For continuous scales, need to back transform from transformed range
	# to original values
	x_trans <- scales$x$trans$inverse(xseq)

Reprex:

library(tidyverse)
fn <- function(x) {exp(x)}
data <- tibble(x = c(1:10), y=exp(x))
data %>% ggplot(aes(x, y)) + geom_point() + stat_function(fun=fn)

data %>% ggplot(aes(x, y)) + geom_point() + stat_function(fun=fn) +
  scale_x_log10()

^{Created on 2020-03-21 by the reprex package (v0.3.0)}

[clauswilke]

I've prepared a PR that fixes this. #3908