fill not quite right in some gf_dist plots

[homerhanumat]

This works well enough:

x <- seq(0.25, 0.75, by = 0.01)
y <- dbeta(x, 2, 4)
df <- data.frame(x, y)
gf_dist("beta", params = list(shape1 = 2, shape2 = 4)) +
  geom_area(data = df,
            mapping = aes(x = x, y = y),
            fill = "burlywood")

But this is visibly off:

gf_dist("beta", params = list(shape1 = 2, shape2 = 4),
            fill = ~(x >= 0.25 & x <= 0.75),
            geom = "area")

[rpruim]

Actually, the second plot is doing just what ggplot2 thinks you are asking for. So the question is whether we should add something to gf_dist() to make the kind of plot you want more easily.

Here's what's going on in the example above: There are only two groups (TRUE and FALSE), and geom_area() doesn't treat the two tail pieces separately unless you make them different groups (you could color them the same with custom scales). So that bit of extra "red" is part of a larger shaded region, most of which is covered by the central piece. If we set alpha lower we can see this more clearly:

library(ggformula)
theme_set(theme_bw())
gf_dist("beta", params = list(shape1 = 2, shape2 = 4),
        fill = ~(x >= 0.25 & x <= 0.75),
        geom = "area", alpha = 0.3) 

One solution is to use cut() to create three groups:

gf_dist("beta", params = list(shape1 = 2, shape2 = 4),
        fill = ~cut(x, c(-Inf, 0.25, 0.75, Inf)),
        geom = "area", alpha = 0.3) 

Created on 2018-09-26 by the reprex package (v0.2.0).

[rpruim]

One way to make things easier might be to add a breaks argument and have the stat create a variable that could be used for fill. You'd still need a custom scale to get the tails to be the same color, however.

I'll have to think about whether (and how best) to implement something like this.

[rpruim]

Side note: you can avoid the use of list() and layer things without reverting to + and direct calls to geoms.

library(ggformula)
theme_set(theme_bw())
gf_dist("beta", shape1 = 2, shape2 = 4,
        fill = ~cut(x, c(-Inf, 0.25, 0.75, Inf)),
        geom = "area", alpha = 0.5)  %>%
  gf_dist("beta", shape1 = 2, shape2 = 4)

Created on 2018-09-26 by the reprex package (v0.2.0).

[homerhanumat]

Thanks, no big hurry, as my students aren't making these graphs yet. reprex seems quite handy!

[rpruim]

Alternatively, you could do it this way. As bonus you get nice labeling. If you want to further modify the plot, be sure to use return = "plot". The default returns the probabilities as a vector, just like pbeta(). This is how I have students build these plots.

library(mosaic)
theme_set(theme_bw())
xpbeta(c(0.25, 0.75), shape1 = 2, shape2 = 4)

#> [1] 0.3671875 0.9843750

Created on 2018-09-27 by the reprex package (v0.2.0).

[rpruim]

I'm going to close this since the behavior is as "expected", ie, just like ggplot2. But I'll open a new issue about possibly adding some functionality to gf_dist() in the future.