Layer data mapping are ignored in stat_function()

[iagogv3]

I have tried to use stat_function in combination with aes(group = ...) and args=list(...), taking into account the documentation

stat_function() understands the following aesthetics (required aesthetics are in bold):
- group
- y

of what I expected different curve plots depending on arguments taken from each group, but I only obtained one curve.

I have asked in stackoverflow if there is a way to do this, but further, I was looking for examples using these aesthetics combined with stat_function, and I did not find anything.

I copy my example of the link:

library(ggplot2)
mtcars$vs <- as.factor(mtcars$vs)
ggplot(mtcars,aes(x=mpg, fill = vs, colour = vs)) + geom_density(alpha = 0.1) + 
    stat_function(aes(group = vs), fun = dnorm, args = list(mean = mean(mtcars$mpg), sd = sd(mtcars$mpg)))

I have also tried deleting fill = vs, colour = vs, and changing it by group = vs, but no result. Since I have also not found any example, not even in the vignette ("ggplot2-specs"), I wonder if this really works.

Thank you!

[batpigandme]

Adding a reprex for clarity

suppressPackageStartupMessages(library(tidyverse))
mtcars$vs <- as.factor(mtcars$vs)
mtcars %>%
  group_by(vs) %>%
  ggplot(aes(x = mpg, fill = vs, colour = vs)) +
  geom_density(alpha = 0.1) +
  stat_function(aes(group = vs), fun = dnorm, args = list(mean = mean(mtcars$mpg), sd = sd(mtcars$mpg)))

^{Created on 2019-06-13 by the reprex package (v0.3.0)}

[paleolimbot]

stat_function() doesn't actually understand the group aesthetic...it is intended to display the result of one function over the x range. If there is more than one group, stat_function() will draw two copies of the result, but never actually looks at the data to get values.

What you are trying to do is more like implementing your own Stat:

library(ggplot2)

StatNormalDensity <- ggproto(
  "StatNormalDensity", Stat,
  required_aes = "x",
  default_aes = aes(y = stat(y)),
  
  compute_group = function(data, scales, xlim = NULL, n = 101) {
    mean_value <- mean(data$x)
    sd_value <- sd(data$x)
    fun <- function(x) dnorm(x, mean = mean_value, sd = sd_value)
    StatFunction$compute_group(data, scales, fun = fun, xlim = xlim, n = n)
  }
)

mtcars$vs <- as.factor(mtcars$vs)
ggplot(mtcars, aes(x = mpg, fill = vs, colour = vs)) +
  geom_density(alpha = 0.1) +
  geom_line(stat = StatNormalDensity, size = 1)

^{Created on 2019-06-13 by the reprex package (v0.2.1)}

From the ggplot2 development perspective, It is assumed that all Stats and Geoms understand the group aesthetic:

ggplot2/R/stat-.r

Lines 146 to 148 in b560662

	aesthetics = function(self) {
	c(union(self$required_aes, names(self$default_aes)), "group")
	}

ggplot2/R/geom-.r

Lines 153 to 155 in b560662

	aesthetics = function(self) {
	c(union(self$required_aes, names(self$default_aes)), self$optional_aes, "group")
	}

Aesthetics will hopefully be better documented after #3293 is addressed, which should solve this issue.

[batpigandme]

Nice. We should probably clarify the docs then. (Maybe re-title this issue, and mark it for tidy-dev-day?)

[paleolimbot]

I think it might also be worth modifying the constructor for stat_function() such that it warns if the user supplies either the mapping or data, since both are ignored anyway. Something like the warning in geom_abline() would work nicely:

ggplot2/R/geom-abline.r

Lines 88 to 93 in b560662

	if (!is.null(mapping)) {
	warn_overwritten_args("geom_abline()", "mapping", c("slope", "intercept"))
	}
	if (!is.null(data)) {
	warn_overwritten_args("geom_abline()", "data", c("slope", "intercept"))
	}

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/