R/AddMissingInidcatorVariables.R

In citronmeliss/predictionpackr: Predictionpackr

#' Add missing indicator variables function
#'
#' Creates missing indicator variables for each predictor variable (feature) with missing values
#' @param study.sample Data Frame. The study sample. No default
#' @param outcome.name String. The name of outcome variable. Defaults to "composite"
#' @param features Character vector. The names of the feature variables. Defaults to NULL, in which case all columns except outcome.name and those potentially listed in excluded.columns are treated as features.
#' @param excluded.columns Character vector. The names of columns that should not be treated as features. Defaults to NULL and is then ignored. Is also ignored if features is specified.
#' @export
AddMissingIndicatorVariables <- function(study.sample, outcome.name = "composite",
                                         features = NULL, excluded.columns = c("tc")) {

  ## Test that all columns specified in outcome.name, features and excluded
  ## features (if any) are in study.data
  column.names <- colnames(study.sample)
  stopifnot(all(c(outcome.name, features, excluded.columns) %in% column.names))
  ## Create vector of features
  if (is.null(features))
    features <- column.names[!(column.names %in% c(outcome.name, excluded.columns))]
  ## Get feature data only
  feature.data <- study.sample[, features]
  ## Define function to get either number of missing values or indicator variable
  GetMissingInfo <- function(x, n = TRUE, variable = FALSE) {
    n.missing <- sum(is.na(x))
    indicator.variable <- NULL
    if (n.missing > 0)
      indicator.variable <- as.numeric(!is.na(x))
    if (n & !variable)
      return(n.missing)
    if (variable)
      return(indicator.variable)
  }
  ## Count number of missing values in each feature
  missing.data.counts <- lapply(feature.data, GetMissingInfo)
  missing.data.matrix <- do.call(rbind, missing.data.counts)
  ## results$missing.data.table <<- create.missing.data.table(missing.data.matrix)
  ## Create a dummy variable for each feature with missing
  indicator.variables <- lapply(feature.data, GetMissingInfo, variable = TRUE)
  indicator.data <- do.call(cbind, indicator.variables)
  colnames(indicator.data) <- paste0(colnames(indicator.data), "_missing")
  study.sample <- data.frame(study.sample, indicator.data)
  return(study.sample)
}