check_predictions() does not work objects of class `_multinom`, created with tidymodels

[davidkane9]

library(tidyverse)
library(tidymodels)
#> Registered S3 method overwritten by 'future':
#>   method               from      
#>   all.equal.connection parallelly
library(easystats)
#> # Attaching packages: easystats 0.7.4 (red = needs update)
#> ✖ bayestestR  0.15.3   ✔ correlation 0.8.7 
#> ✖ datawizard  1.0.2    ✖ effectsize  1.0.0 
#> ✖ insight     1.2.0    ✖ modelbased  0.11.0
#> ✖ performance 0.13.0   ✖ parameters  0.25.0
#> ✔ report      0.6.1    ✔ see         0.11.0
#> 
#> Restart the R-Session and update packages with `easystats::easystats_update()`.

fit_obj <- multinom_reg() |>
  fit(species ~ bill_length_mm, data = penguins)

check_predictions(fit_obj)
#> Error: Could not simulate responses. Maybe there is no `simulate()` for objects
#>   of class `_multinom`?

check_predictions(extract_fit_engine(fit_obj))
#> Error: Could not simulate responses. Maybe there is no `simulate()` for objects
#>   of class `multinom`?

^{Created on 2025-06-23 with reprex v2.1.1}

[etiennebacher]

(Probably) Related to the broader issue on the support for tidymodels: #165

[strengejacke]

This is, as the error states, in "issue" with the nnet package, which provides no simulate() methods for multinom models:

library(easystats)

fit_obj <- nnet::multinom(species ~ bill_len, data = penguins)
check_predictions(fit_obj)
#> Error: Could not simulate responses. Maybe there is no `simulate()` for objects
#>   of class `multinom`?

simulate(fit_obj)
#> Error in UseMethod("simulate"): no applicable method for 'simulate' applied to an object of class "c('multinom', 'nnet')"

^{Created on 2025-06-23 with reprex v2.1.1}

Unless you go Bayesian, there's nothing we can do about this.

[davidkane9]

Thanks for the quick replies! Is there a multinomial model function/package which works well check_predictions?

I am not committed to tidymodels for this application, although I certainly would love to see tidymodels/easystats work even better together!

[strengejacke]

I think brglm2::bracl() works, although there's a warning about factor levels. We may need to check if the output is correct.