R/learner_obliqueRSF_surv_obliqueRSF.R

In mlr-org/mlr3extralearners: Extra Learners For mlr3

#' @title Survival Oblique Random Survival Forest Learner
#' @name mlr_learners_surv.obliqueRSF
#' @author adibender
#'
#' @description
#' Oblique random forest.
#' Calls [obliqueRSF::ORSF()] from \CRANpkg{obliqueRSF}.
#' Note that `obliqueRSF` has been superseded by `aorsf.`
#' We highly recommend you use aorsf to fit oblique random survival forests: see https://github.com/bcjaeger/aorsf or install from CRAN with install.packages('aorsf').
#' @template learner
#' @templateVar id surv.obliqueRSF
#'
#' @section Custom mlr3 parameters:
#' - `mtry`:
#'   - This hyperparameter can alternatively be set via the added hyperparameter `mtry_ratio`
#'     as `mtry = max(ceiling(mtry_ratio * n_features), 1)`.
#'     Note that `mtry` and `mtry_ratio` are mutually exclusive.
#'
#' @section Initial parameter values:
#' - `verbose` is initialized to `FALSE`
#'
#' @references
#' `r format_bib("jaeger_2019")`
#'
#' @template seealso_learner
#' @template example
#' @export
LearnerSurvObliqueRSF = R6Class("LearnerSurvObliqueRSF",
  inherit = mlr3proba::LearnerSurv,

  public = list(
    #' @description
    #' Creates a new instance of this [R6][R6::R6Class] class.
    initialize = function() {
      ps = ps(
        alpha = p_dbl(default = 0.5, tags = "train"),
        ntree = p_int(default = 100L, lower = 1L, tags = "train"),
        eval_times = p_uty(tags = "train"),
        min_events_to_split_node = p_int(default = 5L, lower = 1L, tags = "train"),
        min_obs_to_split_node = p_int(default = 10L, lower = 1L, tags = "train"),
        min_obs_in_leaf_node = p_int(default = 5L, lower = 1L, tags = "train"),
        min_events_in_leaf_node = p_int(default = 1L, lower = 1L, tags = "train"),
        nsplit = p_int(default = 25L, lower = 1, tags = "train"),
        gamma = p_dbl(default = 0.5, lower = 1e-16, tags = "train"),
        max_pval_to_split_node = p_dbl(lower = 0, upper = 1, default = 0.5,
          tags = "train"),
        mtry = p_int(lower = 1, tags = "train"),
        mtry_ratio = p_dbl(0, 1, tags = "train"),
        dfmax = p_int(lower = 1, tags = "train"),
        use.cv = p_lgl(default = FALSE, tags = "train"),
        verbose = p_lgl(default = TRUE, tags = "train"),
        compute_oob_predictions = p_lgl(default = FALSE, tags = "train"),
        random_seed = p_int(tags = "train")
      )

      ps$values = list(verbose = FALSE)

      super$initialize(
        id = "surv.obliqueRSF",
        packages = c("mlr3extralearners", "obliqueRSF", "pracma"),
        feature_types = c("integer", "numeric", "factor", "ordered"),
        predict_types = c("crank", "distr"),
        param_set = ps,
        properties = c("missings", "oob_error"),
        man = "mlr3extralearners::mlr_learners_surv.obliqueRSF",
        label = "Oblique Random Forest"
      )
    },

    #' @description
    #' Integrated brier score OOB error extracted from the model slot `oob_error`.
    #' Concordance is also available.
    #' @return `numeric()`.
    oob_error = function() {
      self$model$oob_error$integrated_briscr[2, ]
    }
  ),

  private = list(
    .train = function(task) {
      pv = self$param_set$get_values(tags = "train")
      pv = convert_ratio(pv, "mtry", "mtry_ratio", length(task$feature_names))
      targets = task$target_names

      invoke(
        obliqueRSF::ORSF,
        data     = data.table::setDF(task$data()),
        time     = targets[1L],
        status   = targets[2L],
        .args    = pv
      )
    },

    .predict = function(task) {

      time = self$model$data[[task$target_names[1]]]
      status = self$model$data[[task$target_names[2]]]
      utime = unique(time[status == 1])

      surv = mlr3misc::invoke(predict,
        self$model,
        newdata = ordered_features(task, self),
        times = utime,
        .args = self$param_set$get_values(tags = "predict"))

      mlr3proba::.surv_return(times = utime, surv = surv)
    }
  )
)

.extralrns_dict$add("surv.obliqueRSF", LearnerSurvObliqueRSF)