R/impute_mode.R

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

#' Impute nominal data using the most common value
#'
#' `step_impute_mode()` creates a *specification* of a recipe step that will
#' substitute missing values of nominal variables by the training set mode of
#' those variables.
#'
#' @inheritParams step_center
#' @param modes A named character vector of modes. This is `NULL` until computed
#'   by [prep()].
#' @param ptype A data frame prototype to cast new data sets to. This is
#'   commonly a 0-row slice of the training set.
#' @template step-return
#' @family imputation steps
#' @export
#' @details
#'
#' `step_impute_mode()` estimates the variable modes from the data used in the
#' `training` argument of [prep()]. [bake()] then applies the new values to new
#' data sets using these values. If the training set data has more than one
#' mode, one is selected at random.
#'
#' As of `recipes` 0.1.16, this function name changed from `step_modeimpute()`
#' to `step_impute_mode()`.
#'
#' # Tidying
#'
#' When you [`tidy()`][tidy.recipe()] this step, a tibble is returned with
#' columns `terms`, `value` , and `id`:
#'
#' \describe{
#'   \item{terms}{character, the selectors or variables selected}
#'   \item{value}{character, the mode value}
#'   \item{id}{character, id of this step}
#' }
#'
#' @template case-weights-unsupervised
#'
#' @examplesIf rlang::is_installed("modeldata")
#' data("credit_data", package = "modeldata")
#'
#' ## missing data per column
#' vapply(credit_data, function(x) mean(is.na(x)), c(num = 0))
#'
#' set.seed(342)
#' in_training <- sample(1:nrow(credit_data), 2000)
#'
#' credit_tr <- credit_data[in_training, ]
#' credit_te <- credit_data[-in_training, ]
#' missing_examples <- c(14, 394, 565)
#'
#' rec <- recipe(Price ~ ., data = credit_tr)
#'
#' impute_rec <- rec %>%
#'   step_impute_mode(Status, Home, Marital)
#'
#' imp_models <- prep(impute_rec, training = credit_tr)
#'
#' imputed_te <- bake(imp_models, new_data = credit_te)
#'
#' table(credit_te$Home, imputed_te$Home, useNA = "always")
#'
#' tidy(impute_rec, number = 1)
#' tidy(imp_models, number = 1)
step_impute_mode <-
  function(
    recipe,
    ...,
    role = NA,
    trained = FALSE,
    modes = NULL,
    ptype = NULL,
    skip = FALSE,
    id = rand_id("impute_mode")
  ) {
    add_step(
      recipe,
      step_impute_mode_new(
        terms = enquos(...),
        role = role,
        trained = trained,
        modes = modes,
        ptype = ptype,
        skip = skip,
        id = id,
        case_weights = NULL
      )
    )
  }

step_impute_mode_new <-
  function(terms, role, trained, modes, ptype, skip, id, case_weights) {
    step(
      subclass = "impute_mode",
      terms = terms,
      role = role,
      trained = trained,
      modes = modes,
      ptype = ptype,
      skip = skip,
      id = id,
      case_weights = case_weights
    )
  }

#' @export
prep.step_impute_mode <- function(x, training, info = NULL, ...) {
  col_names <- recipes_eval_select(x$terms, training, info)

  wts <- get_case_weights(info, training)
  were_weights_used <- are_weights_used(wts, unsupervised = TRUE)
  if (isFALSE(were_weights_used)) {
    wts <- NULL
  }

  modes <- vapply(training[, col_names], mode_est, c(mode = ""), wts = wts)
  ptype <- vec_slice(training[, col_names], 0)
  step_impute_mode_new(
    terms = x$terms,
    role = x$role,
    trained = TRUE,
    modes = modes,
    ptype = ptype,
    skip = x$skip,
    id = x$id,
    case_weights = were_weights_used
  )
}

#' @export
bake.step_impute_mode <- function(object, new_data, ...) {
  col_names <- names(object$modes)
  check_new_data(col_names, object, new_data)

  for (col_name in col_names) {
    if (!anyNA(new_data[[col_name]])) {
      next
    }
    if (is.null(object$ptype)) {
      cli::cli_warn(
        c(
          "!" = "{.arg ptype} was added to {.fn step_impute_mode} after this \\
              recipe was created.",
          "i" = "Regenerate your recipe to avoid this warning."
        )
      )
    } else {
      new_data[[col_name]] <- vctrs::vec_cast(
        new_data[[col_name]],
        object$ptype[[col_name]]
      )
    }
    mode_val <- cast(object$modes[[col_name]], new_data[[col_name]])
    new_data[is.na(new_data[[col_name]]), col_name] <- mode_val
  }

  new_data
}

#' @export
print.step_impute_mode <-
  function(x, width = max(20, options()$width - 30), ...) {
    title <- "Mode imputation for "
    print_step(
      names(x$modes),
      x$terms,
      x$trained,
      title,
      width,
      case_weights = x$case_weights
    )
    invisible(x)
  }

mode_est <- function(x, wts = NULL, call = caller_env(2)) {
  if (!is.character(x) & !is.factor(x))
    cli::cli_abort(
      "The data should be character or factor to compute the mode. \\
      Not {.obj_type_friendly {x}}.",
      call = call
    )
  tab <- weighted_table(x, wts = wts)
  modes <- names(tab)[tab == max(tab)]
  sample(modes, size = 1)
}

#' @rdname tidy.recipe
#' @export
tidy.step_impute_mode <- function(x, ...) {
  if (is_trained(x)) {
    res <- tibble(
      terms = names(x$modes),
      value = unname(x$modes)
    )
  } else {
    term_names <- sel2char(x$terms)
    res <- tibble(terms = term_names, value = na_chr)
  }
  res$id <- x$id
  res
}