cant combine group with color or fill aesthetics for geom_density

[ColeGC]

When using the group aesthetic in combination with either fill or color aesthetic, the fill/color aesthetic appears to be ignored (with geom_density and geom_histogram).

The color/fill aesthetic is recognized and behaves as expected if the group aesthetic is also used for faceting. Using the color and fill aesthetic in combination works as expected. Specifying color + fill + group together in combination results in the aesthetic that is not shared with group being ignored.

library(ggplot2)

n_sims <- 10
n_obs <- 100
delta <- 1
tbl <- data.frame("iter" = rep(rep(paste0("s", 1:n_sims), each = n_obs), times = 2),
                  "type" = rep(c("base", "compare"), each = n_obs*n_sims),
                  "val" = c(rnorm(n_sims*n_obs, mean = 0, sd = 1), rnorm(n_sims*n_obs, mean = delta, sd = 1)))

# Want to plot density of 'val', for each type-iter combination; with some coloring to indicate 'type'
# Using group with fill or color aesthetic leads to only group being used (for both geom_density and geom_histogram)
ggplot(tbl, aes(x = val, fill = type, group = iter)) + 
  geom_density()

ggplot(tbl, aes(x = val, color = type, group = iter)) + 
  geom_density()

ggplot(tbl, aes(x = val, fill = type, group = iter)) + 
  geom_histogram()
#> `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

# Using fill + color aesthetics works
ggplot(tbl, aes(x = val, fill = type, color = iter)) + 
  geom_density(alpha = 0.25)

# Using all group, fill, and color aesthetics leads to the aesthetic not shared with group being
ggplot(tbl, aes(x = val, fill = type, color = iter, group = iter)) + 
  geom_density(alpha = 0.25)

ggplot(tbl, aes(x = val, fill = iter, color = type, group = iter)) + 
  geom_density(alpha = 0.25)

ggplot(tbl, aes(x = val, fill = type, color = iter, group = type)) + 
  geom_density(alpha = 0.25)

ggplot(tbl, aes(x = val, fill = iter, color = type, group = type)) + 
  geom_density(alpha = 0.25)

# Using fill + group aeasthetics works when combined with facet_grid
ggplot(tbl, aes(x = val, fill = type, group = iter)) + 
  geom_density(alpha = 0.25) + 
  facet_grid(type ~ ., scales = "fixed")

[clauswilke]

The fundamental problem is that in all cases where fill or color disappears, the code is actually ill-specified. This leads to the question why there isn't a warning. The answer is that we've had a PR pending for a long time but never moved on it: #3251

Maybe we should pick it back up for past 3.3.0? @yutannihilation @thomasp85?

[clauswilke]

Related issue: #3250

[thomasp85]

Sure

[yutannihilation]

Sounds good. Can we add a milestone for the next version after 3.3.0? I think I saw @clauswilke suggested this on some issue, but forgot to comment that I agree.

[clauswilke]

I created a milestone, tentatively called 3.4.0.

[yutannihilation]

Thanks!

[yutannihilation]

Another example from #4143. I think both legends should reflect colour.

library(ggplot2)
library(patchwork)

set.seed(1)
n_sims <- 10
n_obs <- 100
delta <- 1
tbl <- data.frame("iter" = rep(rep(paste0("s", 1:n_sims), each = n_obs), times = 2),
                  "type" = rep(c("base", "compare"), each = n_obs*n_sims),
                  "val" = c(rnorm(n_sims*n_obs, mean = 0, sd = 1), rnorm(n_sims*n_obs, mean = delta, sd = 1)))

p1 <- ggplot(tbl, aes(x = val, fill = type)) + 
  geom_density(aes(colour = type), alpha = 0.25) +
  geom_density(alpha = 0.25)

p2 <- ggplot(tbl, aes(x = val, fill = type)) + 
  geom_density(alpha = 0.25) +
  geom_density(aes(colour = type), alpha = 0.25)


p1 / p2

^{Created on 2020-07-25 by the reprex package (v0.3.0)}

[thomasp85]

@yutannihilation your last example is "correct" since the keys are stacked in order of appearance, so you have one key with coloured strokes below one with black strokes