Naive Bayes slow prediction

[jxu]

I'm not sure if this is a problem with klaR or parsnip per se, but naive bayes prediction takes a long time on a large dataset, much longer than training. It might be because of the large number of warnings generated?

> warnings()
Warning messages:
1: In FUN(X[[i]], ...) :
  Numerical 0 probability for all classes with observation 1
2: In FUN(X[[i]], ...) :
  Numerical 0 probability for all classes with observation 2
3: In FUN(X[[i]], ...) :
  Numerical 0 probability for all classes with observation 3

etc.

[EmilHvitfeldt]

Hello @jxu 👋

This is an known issue, and not much we can do on our end. We do have another engine naivebayes that has a significantly faster prediction time

library(parsnip)
library(modeldata)
library(discrim)

nb_klaR <- naive_Bayes() |>
  set_engine("klaR") |>
  fit(Street ~ ., data = ames)

tictoc::tic()
predict(nb_klaR, ames)
#> observation 2930
#> # A tibble: 2,930 × 1
#>    .pred_class
#>    <fct>      
#>  1 Pave       
#>  2 Pave       
#>  3 Pave       
#>  4 Pave       
#>  5 Pave       
#>  6 Pave       
#>  7 Pave       
#>  8 Pave       
#>  9 Pave       
#> 10 Pave       
#> # ℹ 2,920 more rows
tictoc::toc()
#> 4.247 sec elapsed

nb_naivebayes <- naive_Bayes() |>
  set_engine("naivebayes") |>
  fit(Street ~ ., data = ames)

tictoc::tic()
predict(nb_naivebayes, ames)
#> # A tibble: 2,930 × 1
#>    .pred_class
#>    <fct>      
#>  1 Pave       
#>  2 Pave       
#>  3 Pave       
#>  4 Pave       
#>  5 Pave       
#>  6 Pave       
#>  7 Pave       
#>  8 Pave       
#>  9 Pave       
#> 10 Pave       
#> # ℹ 2,920 more rows
tictoc::toc()
#> 0.025 sec elapsed

[jxu]

Thanks for getting back to me. Is there a reason why naivebayes is not the default? The Ames dataset, 2930 x 74, is not even that big, and naive bayes is supposed to be extremely fast.

[jxu]

Related #1007

[EmilHvitfeldt]

{klaR} was the original engine for naive_Bayes() and when then {naivebayes} engine as added, {klaR} was kept the default to keep the old behavior. E.I. it was likely that some some people relied on the fact that a {klaR} model came through and it would break things if that changed.

tidymodels/discrim@b27a1c5

[simonpcouch]

Will go ahead and close. Glad we've documented this here!

[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.