Aesthetics for position adjustments

[teunbrand]

This PR aims to fix #3026, fix #5445 and fix #3022.

Briefly, it adds the ability for position adjustments to declare their own aesthetics. It is currently only implemented for position_nudge().

On any layer where position = "nudge", one can use nudge_x and nudge_y either as a mapped aesthetic inside aes(), or static aesthetic passed through layer(params).

A few examples:

devtools::load_all("~/packages/ggplot2")
#> ℹ Loading ggplot2

df <- data.frame(x = 1:2, y = 1:2, z = c(0.5, 1))

ggplot(df, aes(x, y)) +
  geom_point(position = "nudge", nudge_x = 0.5)

ggplot(df, aes(x, y, nudge_y = z)) +
  geom_point(position = "nudge")

^{Created on 2024-09-12 with reprex v2.1.1}

[clauswilke]

@teunbrand As you're working on this, I'd suggest you also look at the long history of issues with misplaced boxplots or violins when categories are missing. I realized long ago that the only way to truly fix this is to somehow have position adjustments be aware of aesthetics. Not sure whether your current work can address this, but conceptually it seems related.

See for example the long discussion here: #3022, and various other issues quoted therein.

Also, here is a reprex that highlights the issue. There is no way I am aware of that can properly format and place boxplots when categories are missing.

library(ggplot2)

# boxes for water polo, netball, and gymnastics have the wrong size
ggplot(ggridges::Aus_athletes, aes(height, sport, fill = sex)) +
  geom_boxplot()

# box for water polo is in the wrong location
ggplot(ggridges::Aus_athletes, aes(height, sport, fill = sex)) +
  geom_boxplot(position = position_dodge(preserve = "single"))

# boxes for water polo, netball, and gymnastics are in the wrong location
ggplot(ggridges::Aus_athletes, aes(height, sport, fill = sex)) +
  geom_boxplot(position = position_dodge2(preserve = "single"))

^{Created on 2024-09-12 with reprex v2.0.2}

Let me know if you think I should open a new issue for this. I don't think we have one currently, but am not entirely certain.

[teunbrand]

Hi @clauswilke, I think that the issue you're highlight is fundamentally the same as #3345. I have pondered them, but haven't been able to come up with a good solution for them (yet). In any case, that issue will not be fixed by this PR.

[clauswilke]

Yes, agreed, it's the same as #3345.

[teunbrand]

On second thought, this PR might be able to fix it if we have something like an order aesthetic for position_dodge().
I'll play around with this for a bit.

[teunbrand]

I've added an order aesthetic to position_dodge() to fix #3022 and #3345.

As an example without order, consider the following case where boxes have partially arbitrary order within their x position:

devtools::load_all("~/packages/ggplot2")
#> ℹ Loading ggplot2

set.seed(0)
df <- data.frame(
  x = rep(rep(LETTERS[1:7], c(3, 2, 2, 2, 1, 1, 1)), each = 10),
  y = rnorm(120),
  group = rep(c("X", "Y", "Z")[c(1:3, 1:2, 1, 3, 2:3, 1:3)], each = 10)
)

ggplot(df, aes(x, y, fill = group)) +
  geom_boxplot(position = position_dodge(preserve = "single"))

Now with the order aesthetic, the boxes are always in the same order regardless of their x position.

last_plot() + aes(order = group)

^{Created on 2024-09-18 with reprex v2.1.1}

[teunbrand]

There isn't anything I'd like to add at this moment, so I'm marking this as ready for review

[thomasp85]

There are some restrictions here that don't apply to the classic aesthetics, right? You can't use after_scale() for instance. Is that somehow taken care of?

[teunbrand]

Any surviving modifiers would be ignored. after_stat() is already evaluated, so that isn't really a concern. after_scale() doesn't really apply as position adjustments are computed before scales are applied. It will be interpreted as there is no such aesthetic, therefore the default value is substituted.

I'm not sure whether this was clear, but this does not replace Geom$use_defaults(), it just evaluates the Position$default_aes really.

[thomasp85]

I can be a bit worried that order is so general that it conflicts with aesthetics from extension packages. Have you looked into that?

[teunbrand]

It is hard to search for, but I can confidently tell you that no package on CRAN has a scale_order_*() function. I've personally never seen an extension package use an order aesthetic.

[thomasp85]

LGTM