breaking change: alpha now impacts the line of stat_ based geoms.

[antoine-lizee]

This has changed following #1371, and I think (personally) that this is the wrong decision.
Pros: it's slightly more 'natural' to have alpha impacting both the line and fill (?)
Cons: the line texture is modified by the alpha parameter in a way that is irreversible, and thus you need a different geom for the common use case (to me) of having a solid line with a transparent filling.
Here is an example:

ggplot(diamonds %>% filter(cut %in% c("Very Good", "Ideal")), aes(x = price, fill = cut)) + theme_bw() + 
  geom_histogram(binwidth = 1000) +
  geom_density(aes(x = price, y = ..count.. * 1000), color = "NA", alpha = 0.3) +
  geom_density(aes(x = price, y = ..count.. * 1000), color = "black", fill = NA) +
  facet_grid(cut ~ ., scales = "free_y", space = "free_y")

I would vote for reverting this behavior, as I'm pretty sure it's breaking more graphing code now than it would if reverted for 2.1.

The downside of the old behavior is that you need to change the color directly if you want something lighter.

[antoine-lizee]

A solve-it-all solution would be to have two different alpha parameters.

[joanwa]

yes, please either revert or two different alpha parameters. This change broke many of my graphs...

[antoine-lizee]

Thanks!

[yutannihilation]

I'm sorry for those who suffered with the change I proposed 😭
Hope someday some great person will discover some great solution which has more easier alpha semantics both for beginners and those who already got used to ggplot2....

[dhimmel]

+1 for separate parameters for line_alpha and fill_alpha (or just alpha as it is currently).

[antoine-lizee]

To answer requests from @dhimmel and others who need lines representing densities with alpha, geom_density is the wrong tool as it is designed to show an area. To get a line, use directly stat_density(geom = "line", alpha = 0.1, ...). Example:

ggplot(diamonds %>% filter(cut %in% c("Very Good", "Ideal")), aes(x = price, fill = cut)) + theme_bw() + 
  geom_histogram(binwidth = 1000) +
  stat_density(aes(x = price, y = ..count.. * 1000), geom = "line", color = "black", alpha = 0.3) +
  facet_grid(cut ~ ., scales = "free_y", space = "free_y")

[dhimmel]

@antoine-lizee thanks for the workaround. I had to add position='identity' in ggplot2::stat_density so the distributions were not stacked.