Standardize classing for grouped resampling

[mikemahoney218]

This PR fixes #318 by standardizing how we subclass when performed grouped resampling. Grouped resampling now:

• Always indicates it's been grouped when printed
• Always includes all the classes of the non-grouped variant
• Always adds a group_rset, and group_x class to the rset object
• Always adds a group_x class to the rsplit object

library(rsample)
data(ames, package = "modeldata")

vf <- vfold_cv(ames)
gvf <- group_vfold_cv(ames, Neighborhood)
vf
#> #  10-fold cross-validation 
#> # A tibble: 10 × 2
#>    splits             id    
#>    <list>             <chr> 
#>  1 <split [2637/293]> Fold01
#>  2 <split [2637/293]> Fold02
#>  3 <split [2637/293]> Fold03
#>  4 <split [2637/293]> Fold04
#>  5 <split [2637/293]> Fold05
#>  6 <split [2637/293]> Fold06
#>  7 <split [2637/293]> Fold07
#>  8 <split [2637/293]> Fold08
#>  9 <split [2637/293]> Fold09
#> 10 <split [2637/293]> Fold10
gvf
#> # Group 28-fold cross-validation 
#> # A tibble: 28 × 2
#>    splits             id        
#>    <list>             <chr>     
#>  1 <split [2907/23]>  Resample01
#>  2 <split [2663/267]> Resample02
#>  3 <split [2929/1]>   Resample03
#>  4 <split [2920/10]>  Resample04
#>  5 <split [2736/194]> Resample05
#>  6 <split [2816/114]> Resample06
#>  7 <split [2837/93]>  Resample07
#>  8 <split [2827/103]> Resample08
#>  9 <split [2922/8]>   Resample09
#> 10 <split [2691/239]> Resample10
#> # … with 18 more rows
class(gvf)
#> [1] "group_vfold_cv" "vfold_cv"       "group_rset"     "rset"          
#> [5] "tbl_df"         "tbl"            "data.frame"
class(gvf$splits[[1]])
#> [1] "group_vfold_split" "vfold_split"       "rsplit"
all(class(vf) %in% class(gvf))
#> [1] TRUE
all(class(vf$splits[[1]]) %in% class(gvf$splits[[1]]))
#> [1] TRUE

mc <- mc_cv(ames)
gmc <- group_mc_cv(ames, Neighborhood)
mc
#> # Monte Carlo cross-validation (0.75/0.25) with 25 resamples  
#> # A tibble: 25 × 2
#>    splits             id        
#>    <list>             <chr>     
#>  1 <split [2197/733]> Resample01
#>  2 <split [2197/733]> Resample02
#>  3 <split [2197/733]> Resample03
#>  4 <split [2197/733]> Resample04
#>  5 <split [2197/733]> Resample05
#>  6 <split [2197/733]> Resample06
#>  7 <split [2197/733]> Resample07
#>  8 <split [2197/733]> Resample08
#>  9 <split [2197/733]> Resample09
#> 10 <split [2197/733]> Resample10
#> # … with 15 more rows
gmc
#> # Grouped Monte Carlo cross-validation (0.75/0.25) with 25 resamples  
#> # A tibble: 25 × 2
#>    splits             id        
#>    <list>             <chr>     
#>  1 <split [2205/725]> Resample01
#>  2 <split [2394/536]> Resample02
#>  3 <split [2292/638]> Resample03
#>  4 <split [2129/801]> Resample04
#>  5 <split [2206/724]> Resample05
#>  6 <split [2230/700]> Resample06
#>  7 <split [2222/708]> Resample07
#>  8 <split [2249/681]> Resample08
#>  9 <split [2125/805]> Resample09
#> 10 <split [2303/627]> Resample10
#> # … with 15 more rows
class(gmc)
#> [1] "group_mc_cv" "mc_cv"       "group_rset"  "rset"        "tbl_df"     
#> [6] "tbl"         "data.frame"
class(gmc$splits[[1]])
#> [1] "grouped_mc_split" "mc_split"         "rsplit"
all(class(mc) %in% class(gmc))
#> [1] TRUE
all(class(mc$splits[[1]]) %in% class(gmc$splits[[1]]))
#> [1] TRUE

bs <- bootstraps(ames)
gbs <- group_bootstraps(ames, Neighborhood)
bs
#> # Bootstrap sampling 
#> # A tibble: 25 × 2
#>    splits              id         
#>    <list>              <chr>      
#>  1 <split [2930/1071]> Bootstrap01
#>  2 <split [2930/1072]> Bootstrap02
#>  3 <split [2930/1100]> Bootstrap03
#>  4 <split [2930/1071]> Bootstrap04
#>  5 <split [2930/1078]> Bootstrap05
#>  6 <split [2930/1082]> Bootstrap06
#>  7 <split [2930/1054]> Bootstrap07
#>  8 <split [2930/1104]> Bootstrap08
#>  9 <split [2930/1077]> Bootstrap09
#> 10 <split [2930/1064]> Bootstrap10
#> # … with 15 more rows
gbs
#> # Group bootstrap sampling 
#> # A tibble: 25 × 2
#>    splits              id         
#>    <list>              <chr>      
#>  1 <split [2901/872]>  Bootstrap01
#>  2 <split [2948/1053]> Bootstrap02
#>  3 <split [2933/885]>  Bootstrap03
#>  4 <split [2947/1206]> Bootstrap04
#>  5 <split [3035/1174]> Bootstrap05
#>  6 <split [2958/1112]> Bootstrap06
#>  7 <split [2914/1066]> Bootstrap07
#>  8 <split [2903/839]>  Bootstrap08
#>  9 <split [2920/1321]> Bootstrap09
#> 10 <split [2823/1532]> Bootstrap10
#> # … with 15 more rows
class(gbs)
#> [1] "group_bootstraps" "bootstraps"       "group_rset"       "rset"            
#> [5] "tbl_df"           "tbl"              "data.frame"
class(gbs$splits[[1]])
#> [1] "group_boot_split" "boot_split"       "rsplit"
all(class(bs) %in% class(gbs))
#> [1] TRUE
all(class(bs$splits[[1]]) %in% class(gbs$splits[[1]]))
#> [1] TRUE

is <- initial_split(ames)
gis <- group_initial_split(ames, Neighborhood)
is
#> <Training/Testing/Total>
#> <2197/733/2930>
gis
#> <Training/Testing/Total>
#> <2187/743/2930>
class(gis)
#> [1] "group_initial_split" "initial_split"       "grouped_mc_split"   
#> [4] "mc_split"            "rsplit"
class(is)
#> [1] "initial_split" "mc_split"      "rsplit"
all(class(is) %in% class(gis))
#> [1] TRUE

vs <- validation_split(ames)
gvs <- group_validation_split(ames, Neighborhood)
vs
#> # Validation Set Split (0.75/0.25)  
#> # A tibble: 1 × 2
#>   splits             id        
#>   <list>             <chr>     
#> 1 <split [2197/733]> validation
gvs
#> # Grouped Validation Set Split (0.75/0.25)  
#> # A tibble: 1 × 2
#>   splits             id        
#>   <list>             <chr>     
#> 1 <split [2177/753]> validation
class(gvs)
#> [1] "group_validation_split" "validation_split"       "group_rset"            
#> [4] "rset"                   "tbl_df"                 "tbl"                   
#> [7] "data.frame"
class(gvs$splits[[1]])
#> [1] "group_val_split" "val_split"       "rsplit"
all(class(vs) %in% class(gvs))
#> [1] TRUE
all(class(vs$splits[[1]]) %in% class(gvs$splits[[1]]))
#> [1] TRUE

^{Created on 2022-06-30 by the reprex package (v2.0.1)}

[mikemahoney218]

This is a bit extraneous for this PR, but as part of a future PR (likely to address #79) I want to try and make sure that group_* variants always have all the same attributes as their non-grouped superclass, so that methods dispatching on the non-grouped variant can always handle the grouped variant as well.

[juliasilge]

I've got one question on printing, and then I was wondering if we could write a test to loop through all the relevant rset types and do this, like you posted:

all(class(mc$splits[[1]]) %in% class(gmc$splits[[1]]))

[juliasilge]

Do you think we should print "Grouped" or "Group"? We have both right now.

[mikemahoney218]

Hmm -- I'm maybe 60/40 in favor of "Group", because the output isn't "Grouped" in a dplyr sense.

Do we want to print either, though? Is it ever important to know in interactive use that these splits were made using groups? Programmatic use will look for the group_* classes.

Between the three, I think my favorite is to always prefix with "Group", but it's an extremely weak opinion.

[juliasilge]

Yeah, let's keep the special printing for now and go with "Group". 👍

[mikemahoney218]

Alright, added tests and changed printing in 43c778e 😄

[juliasilge]

Perfect! 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.