brian-j-smith/machineshop

MachineShop: R package of models and tools for machine learning

machineshop's Issues

Documentation about outer loop in nested resampling

Maybe I missed it, but I could not find documentation about how the outer loop is being performed when nested resampling is done. I assume the inner loop is defined via one of the algorithms in section 9.1 in the user guide, e.g. via BootOptimismControl() or CVControl(). How can the outer loop be controlled or looked up?

BootOptimismControl not working on Linux

The following code that uses the optimism-corrected bootstrap resampling works on Windows and Mac OS, but not on Linux. Any idea what is going on (fyi, using CVControl() works under Linux)?

library(MachineShop, warn.conflicts = FALSE)
tuned_model <- TunedModel( 
  XGBTreeModel, 
  grid =  TuningGrid(), 
  control = BootOptimismControl()
)
MachineShop::fit(mpg ~ ., data = mtcars, model = tuned_model)
#> Error in .fit_optim(object, ...): Resampling failed for all models.
#> XGBTreeModel.1: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.2: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.3: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.4: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.5: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.6: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.7: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.8: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")
#> XGBTreeModel.9: `...` must be empty.
#> ✖ Problematic argument:
#> • group = case_comp_name(df, "groups")

sessionInfo()
#> R version 4.1.2 (2021-11-01)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: TUXEDO OS 2
#> 
#> Matrix products: default
#> BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.10.0
#> LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
#> 
#> locale:
#>  [1] LC_CTYPE=de_DE.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=de_DE.UTF-8        LC_COLLATE=de_DE.UTF-8    
#>  [5] LC_MONETARY=de_DE.UTF-8    LC_MESSAGES=de_DE.UTF-8   
#>  [7] LC_PAPER=de_DE.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=de_DE.UTF-8 LC_IDENTIFICATION=C       
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> other attached packages:
#> [1] MachineShop_3.6.2
#> 
#> loaded via a namespace (and not attached):
#>  [1] tidyr_1.3.0         splines_4.1.2       foreach_1.5.2      
#>  [4] prodlim_2023.08.28  stats4_4.1.2        coin_1.4-2         
#>  [7] yaml_2.3.7          progress_1.2.2      globals_0.16.2     
#> [10] ipred_0.9-14        pillar_1.9.0        lattice_0.20-45    
#> [13] glue_1.6.2          digest_0.6.33       hardhat_1.3.0      
#> [16] sandwich_3.0-2      colorspace_2.1-0    recipes_1.0.8      
#> [19] htmltools_0.5.6     Matrix_1.6-1        timeDate_4022.108  
#> [22] pkgconfig_2.0.3     DiceDesign_1.9      listenv_0.9.0      
#> [25] purrr_1.0.2         mvtnorm_1.2-3       scales_1.2.1       
#> [28] gower_1.0.1         lava_1.7.2.1        timechange_0.2.0   
#> [31] tibble_3.2.1        generics_0.1.3      ggplot2_3.4.3      
#> [34] party_1.3-13        TH.data_1.1-2       withr_2.5.0        
#> [37] furrr_0.3.1         nnet_7.3-17         cli_3.6.1          
#> [40] strucchange_1.5-3   survival_3.2-13     magrittr_2.0.3     
#> [43] crayon_1.5.2        polspline_1.1.23    evaluate_0.21      
#> [46] fs_1.6.3            future_1.33.0       fansi_1.0.4        
#> [49] parallelly_1.36.0   MASS_7.3-55         dials_1.2.0        
#> [52] class_7.3-20        tools_4.1.2         data.table_1.14.8  
#> [55] prettyunits_1.1.1   hms_1.1.3           multcomp_1.4-25    
#> [58] lifecycle_1.0.3     matrixStats_1.0.0   kernlab_0.9-32     
#> [61] rsample_1.2.0       munsell_0.5.0       reprex_2.0.2       
#> [64] compiler_4.1.2      rlang_1.1.1         grid_4.1.2         
#> [67] iterators_1.0.14    rstudioapi_0.15.0   rmarkdown_2.24     
#> [70] gtable_0.3.4        codetools_0.2-18    abind_1.4-5        
#> [73] R6_2.5.1            zoo_1.8-12          lubridate_1.9.2    
#> [76] knitr_1.44          dplyr_1.1.3         fastmap_1.1.1      
#> [79] future.apply_1.11.0 utf8_1.2.3          libcoin_1.0-9      
#> [82] modeltools_0.2-23   parallel_4.1.2      Rcpp_1.0.11        
#> [85] vctrs_0.6.3         rpart_4.1.16        tidyselect_1.2.0   
#> [88] xfun_0.40

^{Created on 2023-09-15 with reprex v2.0.2}

Update to use print_step() instead of printer() in print methods

Hello 👋

In the most recent CRAN release of {recipes}, we took the first step towards updating the printing infrastructure to use {cli}. This means that in order for your steps to print properly when the next version of {recipes} gets on CRAN, you will need to have adopted this change as well. Failure to do so won't result in errors, only malformed printing. The change itself is fairly minimal. Please see the following PR tidymodels/recipes#871 for examples of how this change is to be done. Please let me know if you have any questions!

See example below:

# Old
print.step_pls <- function(x, width = max(20, options()$width - 35), ...) {
  cat("PLS feature extraction with ")
  printer(x$columns, x$terms, x$trained, width = width)
  invisible(x)
}

# New
print.step_pls <- function(x, width = max(20, options()$width - 35), ...) {
  title <- "PLS feature extraction with "
  print_step(x$columns, x$terms, x$trained, width = width, title = title)
  invisible(x)
}

prep.ModelRecipe() skips prep.recipe()

Hi there! We've made a few changes in recipes to fully support modern tidyselect. You can see that PR here
tidymodels/recipes#595

In the process, we slightly tweaked bake() to more directly use an element of a prepped recipe called $last_term_info. This element is only added to the recipe at prep() time, and is something we now left_join() against in bake().

It seems that prep.ModelRecipe() can skip the call to prep() if the recipe looks fully trained.

MachineShop/R/ModelRecipe.R

Lines 76 to 85 in 8777d6b

	prep.ModelRecipe <- function(x, ...) {
	if (!recipes::fully_trained(x)) {
	x <- new(class(x), prep(as(x, "recipe"), retain = FALSE))
	case_name_var <- "(names)"
	x$template[[case_name_var]] <- as.character(x$template[[case_name_var]])
	x$orig_lvls[[case_name_var]] <- list(values = NA, ordered = NA)
	x$levels[[case_name_var]] <- x$orig_lvls[[case_name_var]]
	}
	x
	}

For better or worse, a recipe looks fully trained if there are 0 steps, which means that prep.recipe() won't be called in those cases - meaning that $last_term_info won't get added onto the recipe. This causes this failure in our revdeps
tidymodels/recipes#595 (comment)

Is there any way you can just let prep.recipe() get called unconditionally in prep.ModelRecipe()?

As an FYI, we plan to submit recipes to CRAN fairly soon.

neural network with mnist

Hi,
The model gives an error in the neural network operation of the mnist data. Could the model not work in the multi-factor target variable?
thanks.

Release MachineShop 2.8.0.9000

Prepare for release:

• Check current CRAN check results
• Polish NEWS
• urlchecker::url_check()
• devtools::check(remote = TRUE, manual = TRUE)
• devtools::check_win_devel()
• rhub::check_for_cran()
• revdepcheck::revdep_check(num_workers = 4)
• Update cran-comments.md
• Draft blog post

Submit to CRAN:

• usethis::use_version('minor')
• devtools::submit_cran()
• Approve email

Wait for CRAN...

• Accepted 🎉
• usethis::use_github_release()
• usethis::use_dev_version()
• Finish blog post
• Tweet
• Add link to blog post in pkgdown news menu

Feature importance p-values

In addition to the permutation-based feature importance, there is permutation-based p-values for the feature importance (Altmann, A., Tolosi, L., Sander, O. & Lengauer, T. (2010). Permutation importance: a corrected feature importance measure, Bioinformatics 26:1340-1347). There is essentially only the ranger package that implements this via the importance_pvalues function. Would you think that such a function is helpful? I could imagine that this may aid in judging whether a feature is relevant or not.

prepper now in recipes

Instead of rsample (to signficantly reduce dependencies)

Different ROC AUC output / using formula and recipe