Inconsistency with return type for probability type predictions for two class GAM models

[oj713]

The problem

I'm encountering an inconsistency with generating probability predictions for two class GAM models. The predictions are being returned as arrays within a tibble, rather than just a tibble of numeric values. I believe this issue arises because mgcv returns an array of predictions, rather than a vector.

Issue #541 covers a similar issue around GAM class predictions where mgcv's prediction format was also the source of the problem. I believe the fix for this problem would closely mirror the fix for that issue (fix #542).

Reproducible example

library(parsnip)

set.seed(33)
twoClassSim <- caret::twoClassSim()

gam <- gen_additive_mod() |>
  set_mode("classification") |>
  fit(Class ~ Linear01, 
      data = twoClassSim)

pred <- predict(gam, head(twoClassSim), type = "prob")

# demonstrating error
class(pred$.pred_Class1)
#> [1] "array"
dplyr::glimpse(pred)
#> Rows: 6
#> Columns: 2
#> $ .pred_Class1 <dbl> <array[6]>
#> $ .pred_Class2 <dbl> <array[6]>

# session information 
sessionInfo()
#> R version 4.2.0 (2022-04-22)
#> Platform: aarch64-apple-darwin20 (64-bit)
#> Running under: macOS Monterey 12.3.1
#> 
#> Matrix products: default
#> BLAS:   /Library/Frameworks/R.framework/Versions/4.2-arm64/Resources/lib/libRblas.0.dylib
#> LAPACK: /Library/Frameworks/R.framework/Versions/4.2-arm64/Resources/lib/libRlapack.dylib
#> 
#> locale:
#> [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] parsnip_0.2.1
#> 
#> loaded via a namespace (and not attached):
#>  [1] Rcpp_1.0.8.3         lubridate_1.8.0      lattice_0.20-45     
#>  [4] listenv_0.8.0        tidyr_1.2.0          class_7.3-20        
#>  [7] digest_0.6.29        ipred_0.9-12         foreach_1.5.2       
#> [10] utf8_1.2.2           parallelly_1.31.1    R6_2.5.1            
#> [13] plyr_1.8.7           stats4_4.2.0         reprex_2.0.1        
#> [16] hardhat_0.2.0        evaluate_0.15        ggplot2_3.3.5       
#> [19] highr_0.9            pillar_1.7.0         rlang_1.0.2         
#> [22] caret_6.0-92         data.table_1.14.2    rstudioapi_0.13     
#> [25] rpart_4.1.16         Matrix_1.4-1         rmarkdown_2.14      
#> [28] splines_4.2.0        gower_1.0.0          stringr_1.4.0       
#> [31] munsell_0.5.0        compiler_4.2.0       xfun_0.30           
#> [34] pkgconfig_2.0.3      mgcv_1.8-40          globals_0.14.0      
#> [37] htmltools_0.5.2      nnet_7.3-17          tidyselect_1.1.2    
#> [40] tibble_3.1.6         prodlim_2019.11.13   codetools_0.2-18    
#> [43] fansi_1.0.3          future_1.25.0        crayon_1.5.1        
#> [46] dplyr_1.0.9          withr_2.5.0          ModelMetrics_1.2.2.2
#> [49] MASS_7.3-56          recipes_0.2.0        grid_4.2.0          
#> [52] nlme_3.1-157         gtable_0.3.0         lifecycle_1.0.1     
#> [55] magrittr_2.0.3       pROC_1.18.0          scales_1.2.0        
#> [58] future.apply_1.9.0   cli_3.3.0            stringi_1.7.6       
#> [61] reshape2_1.4.4       fs_1.5.2             timeDate_3043.102   
#> [64] ellipsis_0.3.2       generics_0.1.2       vctrs_0.4.1         
#> [67] lava_1.6.10          iterators_1.0.14     tools_4.2.0         
#> [70] glue_1.6.2           purrr_0.3.4          parallel_4.2.0      
#> [73] fastmap_1.1.0        survival_3.3-1       yaml_2.3.5          
#> [76] colorspace_2.0-3     knitr_1.39

^{Created on 2022-04-28 by the reprex package (v2.0.1)}

[oj713]

I forked the parsnip repository and created a possible fix, available here.

[EmilHvitfeldt]

Hello @oj713 That looks great, good find! If you want to, we would be happy if you would want to create a PR, if not them I'll take it from here.

[oj713]

Hi @EmilHvitfeldt, thank you! I'll go ahead and create a PR now.

[EmilHvitfeldt]

Closed via #709

[github-actions]

This issue 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.