R/filter_extreme_rules.R

In CRE: Interpretable Discovery and Inference of Heterogeneous Treatment Effects

#' @title
#' Filter extreme decision rules
#'
#' @description
#' Discards rules with too few or too many observations.
#'
#' @param rules_matrix A rules matrix.
#' @param rules_list A list of rules (names).
#' @param t_ext A threshold in [0, 0.5) range to truncate too generic or too
#' specific rules.
#'
#' @keywords internal
#'
#' @return
#' A rules matrix without the rare/common rules.
#'
filter_extreme_rules <- function(rules_matrix, rules_list, t_ext) {

  logger::log_debug("Filtering extreme rules...")

  # Identify rules with too few or too many observations
  ind <- 1:dim(rules_matrix)[2]
  sup <- apply(rules_matrix, 2, mean)
  elim <- which((sup < t_ext) | (sup > (1 - t_ext)))
  if (length(elim) > 0) ind <- ind[-elim]

  # Remove rules with too few/too many observations
  rules_matrix <- rules_matrix[, ind, drop = FALSE]
  rules_list <- rules_list[ind]
  colnames(rules_matrix) <- rules_list

  logger::log_debug("Done with filtering extreme rules.")

  return(rules_matrix)
}