Support multidimensional dists / rvars in point_interval

[mjskay]

Pinging off of #14 (comment)

Probably add a column for the index

[mitchelloharawild]

I've pushed an update to {fabletools} onto CRAN yesterday which prepares for the next release of {distributional}. It doesn't use {ggdist} for the graphics yet as the existing functionality being replaced supports multivariate distributions. If you'd like to push {ggdist} to CRAN soon, then I can send {distributional} to CRAN.

I think this is the last issue preventing {fabletools} from using {ggdist}, and removing the {ggplot2} import from {distributional}. If it could be included in the next release that would allow me to swap over to {ggdist} sooner, but if not I still think releasing the updates to {ggdist} and {distributional} is worthwhile now.

[mjskay]

Cool sounds good, I'll push a minor release of {ggdist} soon to ensure forward compatibility with {distributional} and then prioritize this issue in the next major release.

[mjskay]

@mitchelloharawild ggdist 3.0.1, which should be forward-compatible with the new {distributional}, is on its way to cran. Please let me know if you run into any issues with it and I will push a new version.

[mjskay]

@mitchelloharawild I'm working through some last issues in prep for a new release and I want to make sure I hit the couple of things you need, including this one. Is there anything else not on this milestone that would be helpful?

[mjskay]

@mitchelloharawild a potential solution to this is now on master: the point_interval() family of functions will now produce a .index column for multivariate distribution objects and rvars. Example:

library(ggplot2)
library(dplyr)
library(distributional)
library(ggdist)

tibble(
  x = c(
    dist_multivariate_normal(list(c(1:4)), list(diag(4))),
    dist_normal()
  ),
  y = c("a","b")
) %>%
  median_qi(x, .width = c(.66, .95))
#> # A tibble: 10 x 8
#>        x y     .index  .lower .upper .width .point .interval
#>    <dbl> <chr>  <int>   <dbl>  <dbl>  <dbl> <chr>  <chr>    
#>  1     1 a          1  0.0458  1.95    0.66 median qi       
#>  2     2 a          2  1.05    2.95    0.66 median qi       
#>  3     3 a          3  2.05    3.95    0.66 median qi       
#>  4     4 a          4  3.05    4.95    0.66 median qi       
#>  5     0 b         NA -0.954   0.954   0.66 median qi       
#>  6     1 a          1 -0.960   2.96    0.95 median qi       
#>  7     2 a          2  0.0400  3.96    0.95 median qi       
#>  8     3 a          3  1.04    4.96    0.95 median qi       
#>  9     4 a          4  2.04    5.96    0.95 median qi       
#> 10     0 b         NA -1.96    1.96    0.95 median qi

^{Created on 2022-01-16 by the reprex package (v2.0.1)}

This can be used with (e.g.) geom_pointinterval() by mapping .index to something:

tibble(
  x = dist_multivariate_normal(list(c(1:4)), list(diag(4))),
  y = c("a","b")
) %>%
  median_qi(x, .width = c(.66, .95)) %>%
  ggplot(aes(x, xmin = .lower, xmax = .upper, y = .index)) +
  geom_pointinterval()

^{Created on 2022-01-16 by the reprex package (v2.0.1)}

Or since stat_pointinterval() uses the point_interval() functions internally, it also generates an .index column that can be mapped after stats are calculated:

tibble(
  x = c(
    dist_multivariate_normal(list(c(1:4)), list(diag(4))),
    dist_normal()
  ),
  y = c("a","b")
) %>%
  ggplot(aes(xdist = x, y = y, group = stat(.index))) +
  stat_pointinterval(position = "dodge")

^{Created on 2022-01-16 by the reprex package (v2.0.1)}

Some limitations:

• currently this only works with intervals; it can't calculate CDFs or densities from multivariate distributions. That is enough that stat/geom_lineribbon, stat/geom_pointinterval, and stat/geom_interval should work at least.
• only quantile intervals currently work for this, not HDIs.

Let me know if this is sufficient for what you need. I think this is the last issue before I draft a new release.