Dummy Variables Creation via Feature Hashing

step_feature_hash creates a a specification of a recipe step that will convert nominal data (e.g. character or factors) into one or more numeric binary columns using the levels of the original data.

step_feature_hash(
  recipe,
  ...,
  role = "predictor",
  trained = FALSE,
  num_hash = 2^6,
  preserve = FALSE,
  columns = NULL,
  skip = FALSE,
  id = rand_id("feature_hash")
)

# S3 method for step_feature_hash
tidy(x, ...)

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 which factor variables will be used to create the dummy variables. See selections() for more details. The selected variables must be factors. For the tidy method, these are not currently used.
role	For model terms created by this step, what analysis role should they be assigned?. By default, the function assumes that the binary dummy variable columns created by the original variables will be used as predictors in a model.
trained	A logical to indicate if the quantities for preprocessing have been estimated.
num_hash	The number of resulting dummy variable columns.
preserve	A single logical; should the selected column(s) be retained (in addition to the new dummy variables)?
columns	A character vector for the selected columns. This is NULL until the step is trained by recipes::prep.recipe().
skip	A logical. Should the step be skipped when the recipe is baked by recipes::bake.recipe()? While all operations are baked when recipes::prep.recipe() 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.
x	A step_feature_hash object.

Value

An updated version of recipe with the new step added to the sequence of existing steps (if any). For the tidy method, a tibble with columns terms (the selectors or original variables selected).

Details

step_feature_hash() will create a set of binary dummy variables from a factor or character variable. The values themselves are used to determine which row that the dummy variable should be assigned (as opposed to having a specific column that the value will map to).

Since this method does not rely on a pre-determined assignment of levels to columns, new factor levels can be added to the selected columns without issue. Missing values result in missing values for all of the hashed columns.

Note that the assignment of the levels to the hashing columns does not try to maximize the allocation. It is likely that multiple levels of the column will map to the same hashed columns (even with small data sets). Similarly, it is likely that some columns will have all zeros. A zero-variance filter (via recipes::step_zv()) is recommended for any recipe that uses hashed columns.

References

Weinberger, K, A Dasgupta, J Langford, A Smola, and J Attenberg. 2009. "Feature Hashing for Large Scale Multitask Learning." In Proceedings of the 26th Annual International Conference on Machine Learning, 1113–20. ACM.

Kuhn and Johnson (2020) Feature Engineering and Selection: A Practical Approach for Predictive Models. CRC/Chapman Hall https://bookdown.org/max/FES/encoding-predictors-with-many-categories.html

See also

recipes::step_dummy(), recipes::step_zv()

Examples

# \donttest{
data(okc, package = "modeldata")

if (is_tf_available()) {
  # This may take a while: 
  rec <- 
    recipe(Class ~ age + location, data = okc) %>%
    step_feature_hash(location, num_hash = 2^6, preserve = TRUE) %>% 
    prep()

  # How many of the 135 locations ended up in each hash column?
  results <- 
    juice(rec, starts_with("location")) %>% 
    distinct() 

  apply(results %>% select(-location), 2, sum) %>% table()
}
#> .
#>  0  1  2  3  4  5  6 
#>  7 15 22 10  5  4  1 
# }