Specify imputation methods for an existing recipe

`hcai-impute` adds various imputation methods to an existing recipe. Currently supports mean (numeric only), new_category (categorical only), bagged trees, or knn.

hcai_impute(
  recipe,
  nominal_method = "new_category",
  numeric_method = "mean",
  numeric_params = NULL,
  nominal_params = NULL
)

Arguments

recipe	A recipe object. imputation will be added to the sequence of operations for this recipe.
nominal_method	Defaults to "new_category". Other choices are "bagimpute", "knnimpute" or "locfimpute".
numeric_method	Defaults to "mean". Other choices are "bagimpute", "knnimpute" or "locfimpute".
numeric_params	A named list with parmeters to use with chosen imputation method on numeric data. Options are bag_model (bagimpute only), bag_trees (bagimpute only), bag_options (bagimpute only), bag_trees (bagimpute only), knn_K (knnimpute only), impute_with (knnimpute only), (bag or knn) or seed_val (bag or knn). See step_bagimpute or step_knnimpute for details.
nominal_params	A named list with parmeters to use with chosen imputation method on nominal data. Options are bag_model (bagimpute only), bag_trees (bagimpute only), bag_options (bagimpute only), bag_trees (bagimpute only), knn_K (knnimpute only), impute_with (knnimpute only), (bag or knn) or seed_val (bag or knn). See step_bagimpute or step_knnimpute for details.

Value

An updated version of `recipe` with the new step added to the sequence of existing steps.

Examples

library(recipes)
#> 
#> Attaching package: ‘recipes’
#> The following object is masked from ‘package:stats’:
#> 
#>     step

n = 100
set.seed(9)
d <- tibble::tibble(patient_id = 1:n,
            age = sample(c(30:80, NA), size = n, replace = TRUE),
            hemoglobin_count = rnorm(n, mean = 15, sd = 1),
            hemoglobin_category = sample(c("Low", "Normal", "High", NA),
                                         size = n, replace = TRUE),
            disease = ifelse(hemoglobin_count < 15, "Yes", "No")
)

# Initialize
my_recipe <- recipe(disease ~ ., data = d)

# Create recipe
my_recipe <- my_recipe %>%
  hcai_impute()
my_recipe
#> Data Recipe
#> 
#> Inputs:
#> 
#>       role #variables
#>    outcome          1
#>  predictor          4
#> 
#> Operations:
#> 
#> Mean Imputation for all_numeric(), -all_outcomes()
#> Filling NA with missing for all_nominal(), -all_outcomes()

# Train recipe
trained_recipe <- prep(my_recipe, training = d)

# Apply recipe
data_modified <- bake(trained_recipe, new_data = d)
missingness(data_modified)
#> # A tibble: 5 x 2
#>   variable            percent_missing
#> * <chr>                         <dbl>
#> 1 patient_id                        0
#> 2 age                               0
#> 3 hemoglobin_count                  0
#> 4 hemoglobin_category               0
#> 5 disease                           0


# Specify methods:
my_recipe <- my_recipe %>%
  hcai_impute(numeric_method = "bagimpute",
    nominal_method = "locfimpute")
my_recipe
#> Data Recipe
#> 
#> Inputs:
#> 
#>       role #variables
#>    outcome          1
#>  predictor          4
#> 
#> Operations:
#> 
#> Mean Imputation for all_numeric(), -all_outcomes()
#> Filling NA with missing for all_nominal(), -all_outcomes()
#> Bagged tree imputation for all_numeric(), -all_outcomes()
#> LOCF Imputation for all_nominal(), -all_outcomes()

# Specify methods and params:
my_recipe <- my_recipe %>%
  hcai_impute(numeric_method = "knnimpute",
    numeric_params = list(knn_K = 4))
#> `knnimpute` depends on another library that does not support character columns yet. If `knnimpute` fails please convert all character columns to factors for knn imputation.
my_recipe
#> Data Recipe
#> 
#> Inputs:
#> 
#>       role #variables
#>    outcome          1
#>  predictor          4
#> 
#> Operations:
#> 
#> Mean Imputation for all_numeric(), -all_outcomes()
#> Filling NA with missing for all_nominal(), -all_outcomes()
#> Bagged tree imputation for all_numeric(), -all_outcomes()
#> LOCF Imputation for all_nominal(), -all_outcomes()
#> K-nearest neighbor imputation for all_numeric(), -all_outcomes()
#> Filling NA with missing for all_nominal(), -all_outcomes()