step_dummy() doesn't capture contrasts

[EmilHvitfeldt]

step_dummy() doesn't capture contrasts, thus won't work bake() if contrasts option if different between prep() and bake() time.

library(recipes)

rec_normal <- recipe(~ ., data = iris) %>%
  step_dummy(all_nominal_predictors()) %>%
  prep()

param <- getOption("contrasts")
go_helmert <- param
go_helmert["unordered"] <- "contr.helmert"
options(contrasts = go_helmert)

rec_helmert <- recipe(~ ., data = iris) %>%
  step_dummy(all_nominal_predictors()) %>%
  prep()

rec_helmert %>%
  bake(iris)
#> # A tibble: 150 × 6
#>    Sepal.Length Sepal.Width Petal.Length Petal.Width Species_X1 Species_X2
#>           <dbl>       <dbl>        <dbl>       <dbl>      <dbl>      <dbl>
#>  1          5.1         3.5          1.4         0.2         -1         -1
#>  2          4.9         3            1.4         0.2         -1         -1
#>  3          4.7         3.2          1.3         0.2         -1         -1
#>  4          4.6         3.1          1.5         0.2         -1         -1
#>  5          5           3.6          1.4         0.2         -1         -1
#>  6          5.4         3.9          1.7         0.4         -1         -1
#>  7          4.6         3.4          1.4         0.3         -1         -1
#>  8          5           3.4          1.5         0.2         -1         -1
#>  9          4.4         2.9          1.4         0.2         -1         -1
#> 10          4.9         3.1          1.5         0.1         -1         -1
#> # ℹ 140 more rows

Everything works up and until now. But rec_normal() was prepped using contr.treatment so it breaks when used in contr.helmet

rec_normal %>%
  bake(iris)
#> Error in `vctrs::vec_locate_matches()`:
#> ! Each value of `needles` must have a match in `haystack`.
#> ✖ Location 5 of `needles` does not have a match.

I could imagine that this will break some production applications, where prepping was done using contr.poly, but the prediction environment (vetiver as an example) would have no idea.

Suggested fix

I think this bug give enough motivation to add a contrast argument to step_dummy() such that:

• it can keep the information
• simplify the code a little
• long term: start deprecation of using global options in steps