Set outline type for geom_density as geom_area

[thomasp85]

The new stroking of geom_ribbon() and geom_area() didn't make it to geom_density() This PR adds it and make it behave like geom_area()

[yutannihilation]

Looks good, but..., I'm wondering if we can safely make this change.

Different from the case of GeomRibbon, I guess there are already many usages of full-stroking outlines on GeomDensity. That's why I didn't include GeomDensity in #3529.

Probably I'm just too timid because changing the outline of GeomDensity is one of my trauma around ggplot2. Let me think a bit more after drinking some coffee...

[yutannihilation]

Suggested change

	outline.type = 'upper') {
	outline.type = "upper") {

[yutannihilation]

You need to add geom_density here to avoid argument checks:

9c21ed7

[thomasp85]

You are already breaking peoples code as geom_density now uses "both" (the default for geom_ribbon()). This will just break the code in a more. beautiful way... 🙂

@hadley do you have any objections to this visual change?

[yutannihilation]

Hmm, you are right. Hope this is OK this time...

[yutannihilation]

@thomasp85 @hadley
Now I feel this is not OK after looking at the examples on geom_density(), a geom with no fill by default. Note that these figures are drawn against master and doesn't reflect this pull request, so it's just my fault in #3529. Sorry I didn't aware of this visual breakage.

I want to propose renaming option "legacy" to "full" and making it default for geom_density() to restore the previous full-stroking outline. What do you think?

[thomasp85]

I am honestly against. The full stroking of density plots has always looked ugly to my eye. I’m for the renaming, though

The sides and lower bound in a density plot is not really data, and stroking them just creates visual noise IMO

[yutannihilation]

The sides and lower bound in a density plot is not really data, and stroking them just creates visual noise IMO

Thanks, agreed. After some investigation, I now agree that drawing the upper lines only is the right thing as base R does so.

plot(density(diamonds$carat, adjust = 5))

What makes me uneasy with upper-only geom_density() is that the line doesn't start from 0 by default, but it seems because geom_density() provides a narrower plot range than plot(density()) (#3387 might be related?). So, you are definitely right in that the side strokes are "visual noise" that have hided this problem. I feel the right one below should be drawn by default. I'll file a new issue for this.

p <- ggplot(diamonds, aes(carat)) +
  geom_density(adjust = 5)

patchwork::wrap_plots(
  p + ggtitle("(default)"),
  p + ggtitle("(expanded)") + expand_limits(x = -0.5)
)

I’m for the renaming, though

Sure.

[thomasp85]

I agree that the rightmost is more pleasing, but I think that ggplot2 has historically erred to the side of keeping the data-range. This is also done by default in violin plots. Extrapolating is something you have to do wilfully, as it creates data from thin air

[yutannihilation]

Sounds fair, but let me dig into the history...