R/pre_process_data.R

In ModelOriented/FairModels: Flexible Tool for Bias Detection, Visualization, and Mitigation

#' Pre-process data
#'
#' Function aggregates all pre-processing algorithms for bias mitigation. User passes unified arguments and specifies type to receive transformed \code{data.frame}
#'
#' @param data \code{data.frame}
#' @param protected factor, protected attribute (sensitive variable) containing information about gender, race etc...
#' @param y numeric, numeric values of predicted variable. 1 should denote favorable outcome.
#' @param type character, type of pre-processing algorithm to be used, one of:
#' \itemize{
#' \item{resample_uniform}
#' \item{resample_preferential}
#' \item{reweight}
#' \item{disparate_impact_remover}
#' }
#' @param ... other parameters passed to pre-processing algorithms
#'
#' @return modified data (\code{data.frame}). In case of type = 'reweight' data has feature `_weights_` containing weights that need to be passed to model.
#' In other cases data is ready to be passed as training data to a model.
#' @export
#'
#' @examples
#' data("german")
#'
#' pre_process_data(german,
#'   german$Sex,
#'   as.numeric(german$Risk) - 1,
#'   type = "disparate_impact_remover",
#'   features_to_transform = "Age"
#' )
pre_process_data <- function(data, protected, y, type = "resample_uniform", ...) {
  switch(type,
    resample_uniform = {
      return(data[resample(protected, y, ...), ])
    },
    resample_preferential = {
      return(data[resample(protected, y, type = "preferential", ...), ])
    },
    reweight = {
      data$`_weights_` <- reweight(protected, y)
      return(data)
    },
    disparate_impact_remover = {
      return(disparate_impact_remover(data, protected, ...))
    }
  )

  # if not in switch:
  stop("type must be equal to one of supported types, see documentation: ?pre_process_data")
}