step_dummy_multi_choice() ignores levels of the input factor

[zenggyu]

The problem

step_dummy_multi_choice() ignores levels of the input factor and only considers levels in the new data. This causes problem if the downstream model uses a dummy variable for a level that does not occur in the new data. This behavior is in contrast to that of step_dummy().

I also notice that the two functions have a common parameter called levels, and it looks like it is intended to explicitly specify the levels to include in the output. However, there are no examples in the documentation and I don't know how to specify (I tried levels = list(x = c("a", "b", "c", "d", "e", "f", "g")) but it failed). Can you add some more explanations on how to use the parameter and/or add an example?

Reproducible example

I would expect the following cases of step_dummy() and step_dummy_multi_choice() to produce the same number of columns. However, step_dummy_multi_choice() does not create columns for levels "a", "b", "f", "g".

By the way, why do step_dummy() and step_dummy_multi_choice() produce different types of output (double vs integer)?

suppressPackageStartupMessages(library(recipes))

# old data
tr <- data.frame(x = factor(c("a", "b", "c"), levels = c("a", "b", "c", "d", "e", "f", "g")))

# new data
te <- data.frame(x = factor(c("c", "d", "e"), levels = c("a", "b", "c", "d", "e", "f", "g")))

tr %>%
  recipe() |>
  step_dummy(x, one_hot = T) |>
  prep() |>
  bake(new_data = te)
#> # A tibble: 3 × 7
#>     x_a   x_b   x_c   x_d   x_e   x_f   x_g
#>   <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1     0     0     1     0     0     0     0
#> 2     0     0     0     1     0     0     0
#> 3     0     0     0     0     1     0     0

tr |>
  recipe() |>
  step_dummy_multi_choice(x) |>
  prep() |>
  bake(new_data = te)
#> # A tibble: 3 × 3
#>     x_c   x_d   x_e
#>   <int> <int> <int>
#> 1     1     0     0
#> 2     0     1     0
#> 3     0     0     1

^{Created on 2022-02-24 by the reprex package (v2.0.0)}

[EmilHvitfeldt]

Hello @zenggyu ! Thanks for filling this issue. You are correct in assuming that step_dummy_multi_choice() should respect the factor levels.

[EmilHvitfeldt]

was closed in #930

[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.