update_role_requirements: Update role specific requirements

In tidymodels/recipes: Preprocessing and Feature Engineering Steps for Modeling

Update role specific requirements

Description

update_role_requirements() allows you to fine tune requirements of the various roles you might come across in recipes (see update_role() for general information about roles). Role requirements can only be altered for roles that exist in the original data supplied to recipe(), they are not applied to columns computed by steps.

Like update_role(), update_role_requirements() is applied to the recipe immediately, unlike the ⁠step_*()⁠ functions which do most of their work at prep() time.

Usage

update_role_requirements(recipe, role, ..., bake = NULL)

Arguments

recipe	A recipe.
role	A string representing the role that you'd like to modify the requirements of. This must be a role that already exists in the recipe.
...	These dots are for future extensions and must be empty.
bake	At bake() time, should a check be done to ensure that all columns of this role that were supplied to recipe() also be present in the new_data supplied to bake()? Must be a single TRUE or FALSE. The default, NULL, won't modify this requirement. The following represents the default bake time requirements of specific types of roles: "outcome": Not required at bake time. Can't be changed. "predictor": Required at bake time. Can't be changed. "case_weights": Not required at bake time by default. NA: Required at bake time by default. Custom roles: Required at bake time by default.

Examples

df <- tibble(y = c(1, 2, 3), x = c(4, 5, 6), var = c("a", "b", "c"))

# Let's assume that you have a `var` column that isn't used in the recipe.
# We typically recommend that you remove this column before passing the
# `data` to `recipe()`, but for now let's pass it through and assign it an
# `"id"` role.
rec <- recipe(y ~ ., df) %>%
  update_role(var, new_role = "id") %>%
  step_center(x)

prepped <- prep(rec, df)

# Now assume you have some "new data" and you are ready to `bake()` it
# to prepare it for prediction purposes. Here, you might not have `var`
# available as a column because it isn't important to your model.
new_data <- df[c("y", "x")]

# By default `var` is required at `bake()` time because we don't know if
# you actually use it in the recipe or not
try(bake(prepped, new_data))

# You can turn off this check by using `update_role_requirements()` and
# setting `bake = FALSE` for the `"id"` role. We recommend doing this on
# the original unprepped recipe, but it will also work on a prepped recipe.
rec <- update_role_requirements(rec, "id", bake = FALSE)
prepped <- prep(rec, df)

# Now you can `bake()` on `new_data` even though `var` is missing
bake(prepped, new_data)

tidymodels/recipes documentation built on Nov. 29, 2024, 3:05 p.m.