Making information available in compute_group for a custom ggplot2::Stat object

[pomchip]

Hi,

I am working on creating a custom geom::Stat and am looking for advises. This new stat needs to operate at the group level, so my ggproto object has a compute_group level. However, in some circumstances, the compute_group function would need to know the range of x and y data in the whole panel (not just the group). What would be the best way to pass this information onto the compute_group function?

Thanks n advance for your suggestions

[technocrat]

Provide for and mapping aes to the un-grouped object is what I’d consider

[pomchip]

@technocrat , are you suggesting to pre-calculate the range of data in each panel and store the information in the data object passed on the ggplot call?

[technocrat]

Something along those lines, like you can overplot different objects within a ggplot object using different data = and aes() mapping

[pomchip]

The purpose of this new stat is to perform the statistical computations internally during the ggplot call... so the suggested approach would defeat the purpose of this function.

[technocrat]

I'm having a difficult time envisioning what the new geom is doing. My thought was that because the ungrouped data frame is in .Global, it can be brought into a geom as the data argument and variables in it subjected to functions and plotted, like this

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(ggplot2)
mtcars$cyl <- as.factor(mtcars$cyl)
mtcars |> 
  group_by(cyl,mpg) |> 
  ggplot(aes(mpg,cyl)) +
    geom_point() +
    geom_line(data = mtcars, aes(y = mean(drat)/drat^2))

^{Created on 2023-03-24 with reprex v2.0.2}

[pomchip]

The issue is with pacetting (especially with free scales), one cannot really rely on the original data object to determine data ranges within a panel.

[technocrat]

Giving a scales geom a function return value based on range() won't work?

[pomchip]

Actually, I found the answer. The scales argument of the compute_group already contains the data ranges.