step_dummy: Create traditional dummy variables

Description

step_dummy() creates a specification of a recipe step that will convert nominal data (e.g. factors) into one or more numeric binary model terms corresponding to the levels of the original data.

Usage

step_dummy(
  recipe,
  ...,
  role = "predictor",
  trained = FALSE,
  one_hot = FALSE,
  contrasts = list(unordered = "contr.treatment", ordered = "contr.poly"),
  preserve = deprecated(),
  naming = dummy_names,
  levels = NULL,
  sparse = "auto",
  keep_original_cols = FALSE,
  skip = FALSE,
  id = rand_id("dummy")
)

Value

An updated version of recipe with the new step added to the sequence of any existing operations.

Arguments

recipe: A recipe object. The step will be added to the sequence of operations for this recipe.
...: One or more selector functions to choose variables for this step. See selections() for more details. The selected variables must be factors.
role: For model terms created by this step, what analysis role should they be assigned? By default, the new columns created by this step from the original variables will be used as predictors in a model.
trained: A logical to indicate if the quantities for preprocessing have been estimated.
one_hot: A logical. For C levels, should C dummy variables be created rather than C-1?
contrasts: A named vector or list of contrast functions names. Defaults to list(unordered = "contr.treatment", ordered = "contr.poly"). If only a single string is passed it will be used for both unordered and ordered.
preserve: This argument has been deprecated. Please use keep_original_cols instead.
naming: A function that defines the naming convention for new dummy columns. See Details below.
levels: A list that contains the information needed to create dummy variables for each variable contained in terms. This is NULL until the step is trained by prep().
sparse: A single string. Should the columns produced be sparse vectors. Can take the values "yes", "no", and "auto". If sparse = "auto" then workflows can determine the best option. Defaults to "auto".
keep_original_cols: A logical to keep the original variables in the output. Defaults to FALSE.
skip: A logical. Should the step be skipped when the recipe is baked by bake()? While all operations are baked when prep() is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using skip = TRUE as it may affect the computations for subsequent operations.
id: A character string that is unique to this step to identify it.

Tidying

When you tidy() this step, a tibble is returned with columns terms, columns , and id:

terms: character, the selectors or variables selected
columns: character, names of resulting columns
id: character, id of this step

Sparse data

This step produces sparse columns if sparse = "yes" is being set. The default value "auto" won't trigger production fo sparse columns if a recipe is prep()ed, but allows for a workflow to toggle to "yes" or "no" depending on whether the model supports sparse_data and if the model is is expected to run faster with the data.

The mechanism for determining how much sparsity is produced isn't perfect, and there will be times when you want to manually overwrite by setting sparse = "yes" or sparse = "no".

Case weights

The underlying operation does not allow for case weights.

Details

step_dummy() will create a set of binary dummy variables from a factor variable. For example, if an unordered factor column in the data set has levels of "red", "green", "blue", the dummy variable bake will create two additional columns of 0/1 data for two of those three values (and remove the original column). For ordered factors, polynomial contrasts are used to encode the numeric values. These defaults are controlled by the contrasts argument. Note that since the contrasts are specified via character strings you will need to have those packages loaded. If you are using this with the tune package, you might need to add that these packages to the pkg option in control_grid().

By default, the excluded dummy variable (i.e. the reference cell) will correspond to the first level of the unordered factor being converted. step_relevel() can be used to create a new reference level by setting the ref_level argument.

This recipe step allows for flexible naming of the resulting variables. For an unordered factor named x, with levels "a" and "b", the default naming convention would be to create a new variable called x_b. The naming format can be changed using the naming argument; the function dummy_names() is the default.

When the factor being converted has a missing value, all of the corresponding dummy variables are also missing. See step_unknown() for a solution.

When data to be processed contains novel levels (i.e., not contained in the training set), a missing value is assigned to the results. See step_other() for an alternative.

If no columns are selected (perhaps due to an earlier step_zv()), bake() will return the data as-is (e.g. with no dummy variables).

Note that, by default, the new dummy variable column names obey the naming rules for columns. If there are levels such as "0", dummy_names() will put a leading "X" in front of the level (since it uses make.names()). This can be changed by passing in a different function to the naming argument for this step.

Also, there are a number of contrast methods that return fractional values. The columns returned by this step are doubles (not integers) when sparse = FALSE. The columns returned when sparse = TRUE are integers.

The package vignette for dummy variables and interactions has more information.

Examples

Run this code

data(Sacramento, package = "modeldata")

# Original data: city has 37 levels
length(unique(Sacramento$city))
unique(Sacramento$city) |> sort()

rec <- recipe(~ city + sqft + price, data = Sacramento)

# Default dummy coding: 36 dummy variables
dummies <- rec |>
  step_dummy(city) |>
  prep()

dummy_data <- bake(dummies, new_data = NULL)

dummy_data |>
  select(starts_with("city")) |>
  glimpse() # level "anything" is the reference level

# Obtain the full set of 37 dummy variables using `one_hot` option
dummies_one_hot <- rec |>
  step_dummy(city, one_hot = TRUE) |>
  prep()

dummy_data_one_hot <- bake(dummies_one_hot, new_data = NULL)

dummy_data_one_hot |>
  select(starts_with("city")) |>
  glimpse() # no reference level

# Obtain the full set of 37 dummy variables using helmert contrasts
dummies_helmert <- rec |>
  step_dummy(city, contrasts = "contr.helmert") |>
  prep()

dummy_data_helmert <- bake(dummies_helmert, new_data = NULL)

dummy_data_helmert |>
  select(starts_with("city")) |>
  glimpse() # no reference level

tidy(dummies, number = 1)
tidy(dummies_one_hot, number = 1)
tidy(dummies_helmert, number = 1)

Run the code above in your browser using DataLab