Add reverse_splits

[mikemahoney218]

This PR fixes #284 by adding a new utility function to "reverse" the analysis and assessment splits.

library(rsample)

(starting_splits <- vfold_cv(mtcars, v = 3))
#> #  3-fold cross-validation 
#> # A tibble: 3 × 2
#>   splits          id   
#>   <list>          <chr>
#> 1 <split [21/11]> Fold1
#> 2 <split [21/11]> Fold2
#> 3 <split [22/10]> Fold3

reverse_splits(starting_splits)
#> #  3-fold cross-validation 
#> # A tibble: 3 × 2
#>   splits          id   
#>   <list>          <chr>
#> 1 <split [11/21]> Fold1
#> 2 <split [11/21]> Fold2
#> 3 <split [10/22]> Fold3

reverse_splits(starting_splits$splits[[1]])
#> <Analysis/Assess/Total>
#> <11/21/32>

reverse_splits(1)
#> Error in `reverse_splits()`:
#> ! `x` must be either an `rsplit` or an `rset` object

reverse_splits(starting_splits, "no dots allowed!")
#> Error in `reverse_splits()`:
#> ! `...` must be empty.
#> ✖ Problematic argument:
#> • ..1 = "no dots allowed!"
#> ℹ Did you forget to name an argument?

^{Created on 2022-06-29 by the reprex package (v2.0.1)}

[mikemahoney218]

Only having two groups here made group_mc_cv() error due to having 0 data in the assessment set. I also did a version of this in #316 to deal with the same issue there (to get both PRs to pass CI).

[juliasilge]

@mattwarkentin would you like to take a look at this?

[mattwarkentin]

I made a couple minor comments but otherwise LGTM! Thanks for addressing this so quickly, @mikemahoney218.

[mattwarkentin]

Oh, one more thought. Have you tested this function with an object from permutations()? Splits for a permutations object have no assessment set by design, so not sure how this should be handled. Maybe trying to reverse the splits for this object should throw an error? Calling assessment() on a permutations object throws an error, so this would be consistent behaviour, I think.

[mikemahoney218]

So this does currently error, though we could possibly provide a friendlier message than what's currently thrown:

library(rsample)

permutations(mtcars, cyl)
#> # Permutation sampling 
#> # Permuted columns: [cyl] 
#> # A tibble: 25 × 2
#>    splits         id            
#>    <list>         <chr>         
#>  1 <split [32/0]> Permutations01
#>  2 <split [32/0]> Permutations02
#>  3 <split [32/0]> Permutations03
#>  4 <split [32/0]> Permutations04
#>  5 <split [32/0]> Permutations05
#>  6 <split [32/0]> Permutations06
#>  7 <split [32/0]> Permutations07
#>  8 <split [32/0]> Permutations08
#>  9 <split [32/0]> Permutations09
#> 10 <split [32/0]> Permutations10
#> # … with 15 more rows

permutations(mtcars, cyl) |> 
  reverse_splits()
#> Error in `rsplit()` at rsample/R/misc.R:23:2:
#> ! At least one row should be selected for the analysis set.

^{Created on 2022-06-29 by the reprex package (v2.0.1)}

(Which, as a sidenote @juliasilge , permutations() isn't in rset_subclasses in compat-vctrs-helpers, otherwise this would cause my test to fail -- should it be added to that list?)

[mattwarkentin]

Related: I am now seeing that the error thrown when calling assessment() on a permutations object is not correct:

library(rsample)

p <- permutations(mtcars, cyl)

assessment(p$splits[[1]])
#> Error in `as.data.frame()`:
#> ! There is no assessment data set for an `rsplit` object with class `rsplit`.

It should read something like:

#> ! There is no assessment data set for an rsplit object from a permutations set.

Or something along those lines...

[mikemahoney218]

Related: I am now seeing that the error thrown when calling assessment() on a permutations object is not correct...

Opened #321 to track that one 😄

[juliasilge]

Thank you so much! 🙌

[github-actions]

This pull request has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.