R/learner_fairml_classif_fairzlrm.R

In mlr3fairness: Fairness Auditing and Debiasing for 'mlr3'

#' @title Classification Fair Logistic Regression With Covariance Constraints Learner
#' @author pfistfl
#' @details 
#' Generalized fair regression model from Zafar et al., 2019 implemented via package `fairml`.
#' The 'unfairness' parameter is set to 0.05 as a default.
#' The optimized fairness metric is statistical parity.
#' 
#' @name mlr_learners_classif.fairzlrm
#'
#' @template class_learner
#' @templateVar id classif.fairzlrm
#' @templateVar caller zlrm
#'
#' @references
#' `r format_bib("zafar19a")`
#' 
#' @template seealso_learner
#' @template example
#' @export
LearnerClassifFairzlrm = R6Class("LearnerClassifFairzlrm",
  inherit = LearnerClassif,

  public = list(
    #' @description
    #' Creates a new instance of this [R6][R6::R6Class] class.
    initialize = function() {
      ps = ps(
        unfairness = p_dbl(lower = 0, upper = 1, tags = "train"),
        intersect = p_lgl(default = TRUE, tags = c("train", "predict"))
      )
      ps$values = list(unfairness = 0.05, intersect = FALSE)
      super$initialize(
        id = "classif.fairzlrm",
        packages = "fairml",
        feature_types = c("integer", "numeric", "factor", "ordered"),
        predict_types = c("response", "prob"),
        properties = "twoclass",
        param_set = ps,
        man = "mlr3fairness::mlr_learners_classif.fairzlrm"
      )
    }
  ),

  private = list(

    .train = function(task) {
      assert_pta_task(task)
      # get parameters for training
      pars = self$param_set$get_values(tags = "train")

      # set column names to ensure consistency in fit and predict
      self$state$feature_names = task$feature_names
      pta = task$col_roles$pta
      r = task$truth()
      s = get_pta(task, intersect = pars$intersect)
      pars = remove_named(pars, "intersect")
      p = task$data(cols = setdiff(task$feature_names, pta))
      p = int_to_numeric(p)
      # use the mlr3misc::invoke function (it's similar to do.call())
      mlr3misc::invoke(fairml::zlrm, response = r, sensitive = s,
        predictors = p, .args = pars)
    },

    .predict = function(task) {
      pta = task$col_roles$pta
      pars = self$param_set$get_values(tags = "predict")
      s = get_pta(task, intersect = pars$intersect)
      p = task$data(cols = setdiff(self$state$feature_names, pta))
      p = int_to_numeric(p)
      if (self$predict_type == "response") {
        pred = mlr3misc::invoke(predict, self$model, new.predictors = p, type = "class")
        list(response = drop(pred))
      } else {
        prob = mlr3misc::invoke(predict, self$model, new.predictors = p, type = "response")
        if (length(task$class_names) == 2L) {
          prob = pprob_to_matrix(prob, task)
        } else {
          prob = prob[, , 1L]
        }
        list(prob = prob)
      }
    }
  )
)