Add randomPlantedForest learners (classif, regr)

DESCRIPTION

@@ -19,7 +19,9 @@ Authors@R: c(
     person("Sebastian", "Fischer", , "sebf.fischer@gmail.com", role = c("cre", "aut")),
     person("Zezhi", "Wang", ,"homura@mail.ustc.edu.cn", role = "ctb"),
     person("John", "Zobolas", ,"bblodfon@gmail.com", role = "ctb",
-           comment = c(ORCID = "0000-0002-3609-8674"))
+           comment = c(ORCID = "0000-0002-3609-8674")),
+    person("Lukas", "Burk", , "burk@leibniz-bips.de", role = "ctb",
+           comment = c(ORCID = "0000-0001-7528-3795"))
   )
 Description: Extra learners for use in mlr3.
 License: LGPL-3
@@ -85,6 +87,7 @@ Suggests:
     prioritylasso (>= 0.3.1),
     pseudo,
     randomForest,
+    randomPlantedForest,
     randomForestSRC,
     ranger,
     remotes,
@@ -108,6 +111,7 @@ Remotes:
     catboost/catboost/catboost/R-package,
     mlr-org/mlr3proba,
     RaphaelS1/survivalmodels,
+    PlantedML/randomPlantedForest,
     xoopR/distr6,
     xoopR/param6,
     xoopR/set6
@@ -119,3 +123,4 @@ NeedsCompilation: no
 Roxygen: list(markdown = TRUE, r6 = TRUE)
 RoxygenNote: 7.2.3
 Config/Needs/website: rmarkdown
+

NAMESPACE

@@ -30,6 +30,7 @@ export(LearnerClassifPART)
 export(LearnerClassifPriorityLasso)
 export(LearnerClassifRandomForest)
 export(LearnerClassifRandomForestSRC)
+export(LearnerClassifRandomPlantedForest)
 export(LearnerDensKDEks)
 export(LearnerDensLocfit)
 export(LearnerDensLogspline)
@@ -65,6 +66,7 @@ export(LearnerRegrRSM)
 export(LearnerRegrRVM)
 export(LearnerRegrRandomForest)
 export(LearnerRegrRandomForestSRC)
+export(LearnerRegrRandomPlantedForest)
 export(LearnerSurvAkritas)
 export(LearnerSurvAorsf)
 export(LearnerSurvBlackBoost)

R/bibentries.R

@@ -24,6 +24,14 @@ bibentries = c( # nolint start
     journal     = "The Annals of Applied Statistics"
   ),
 
+  hiabu_2023 = bibentry("article",
+    title   = "Random Planted Forest: a directly interpretable tree ensemble",
+    author  = "Hiabu, Munir and Mammen, Enno and Meyer, Joseph T.",
+    journal = "arXiv preprint arXiv:2012.14563",
+    doi     = "10.48550/ARXIV.2012.14563",
+    year    = "2023"
+  ),
+
   hothorn_2015 = bibentry("article",
     author      = "Torsten Hothorn and Achim Zeileis",
     title       = "partykit: A Modular Toolkit for Recursive Partytioning in R",
@@ -574,4 +582,4 @@ bibentries = c( # nolint start
     month = "01",
     journal = "University of California, Berkeley"
   )
-) # nolint end
+) # nolint end

R/learner_randomPlantedForest_classif_rpf.R

@@ -0,0 +1,126 @@
+#' @title Classification Random Planted Forest Learner
+#' @author jemus42
+#' @name mlr_learners_classif.rpf
+#'
+#' @description
+#' Random Planted Forest: A directly interpretable tree ensemble.
+#'
+#' Calls [randomPlantedForest::rpf()] from 'randomPlantedForest'.
+#'
+#' @section Initial parameter values:
+#' - `loss`:
+#'   - Actual default: `"L2"`.
+#'   - Initial value: `"exponential"`.
+#'   - Reason for change: Using `"L2"` (or `"L1"`) loss does not guarantee predictions are valid
+#'     probabilities and more akin to the linear predictor of a GLM.
+
+#' @section Custom mlr3 parameters:
+#' - `max_interaction`:
+#'   - This hyperparameter can alternatively be set via `max_interaction_ratio` as
+#'     `max_interaction = max(ceiling(max_interaction_ratio * n_features), 1)`.
+#'     The parameter `max_interaction_limit` can optionally be set as an upper bound, such that
+#'     `max_interaction_ratio * min(n_features, max_interaction_limit)` is used instead.
+#'     This is analogous to `mtry.ratio` in [`classif.ranger`][mlr3learners::mlr_learners_classif.ranger], with
+#'     `max_interaction_limit` as an additional constraint.
+#'     The parameter `max_interaction_limit` is initialized to `Inf`.
+#'
+#' @templateVar id classif.rpf
+#' @template learner
+#'
+#' @section Installation:
+#' Package 'randomPlantedForest' is not on CRAN and has to be installed from GitHub via
+#' `remotes::install_github("PlantedML/randomPlantedForest")`.
+#'
+#' @references
+#' `r format_bib("hiabu_2023")`
+#'
+#' @template seealso_learner
+#' @template example
+#' @export
+LearnerClassifRandomPlantedForest = R6Class("LearnerClassifRandomPlantedForest",
+  inherit = LearnerClassif,
+  public = list(
+    #' @description
+    #' Creates a new instance of this [R6][R6::R6Class] class.
+    initialize = function() {
+      param_set = ps(
+        max_interaction       = p_int(lower = 0, upper = Inf, default = 1, tags = "train"),
+        max_interaction_ratio = p_dbl(lower = 0, upper = 1, tags = "train"),
+        max_interaction_limit = p_int(lower = 1, upper = Inf, tags = c("required", "train")),
+        ntrees                = p_int(lower = 1, upper = Inf, default = 50, tags = "train"),
+        splits                = p_int(lower = 1, upper = Inf, default = 30, tags = "train"),
+        split_try             = p_int(lower = 1, upper = Inf, default = 10, tags = "train"),
+        t_try                 = p_dbl(lower = 0, upper = 1, default = 0.4, tags = "train"),
+        loss                  = p_fct(c("L1", "L2", "logit", "exponential"), default = "L2", tags = "train"),
+        delta                 = p_dbl(lower = 0, upper = 1, default = 1, tags = "train"),
+        epsilon               = p_dbl(lower = 0, upper = 1, default = 0.1, tags = "train"),
+        deterministic         = p_lgl(default = FALSE, tags = "train"),
+        nthreads              = p_int(lower = 1, upper = Inf, default = 1, tags = c("train", "threads")),
+        cv                    = p_lgl(default = FALSE, tags = "train"),
+        purify                = p_lgl(default = FALSE, tags = "train")
+      )
+
+      param_set$values = list(loss = "exponential", max_interaction_limit = Inf)
+
+      super$initialize(
+        id = "classif.rpf",
+        packages = "randomPlantedForest",
+        feature_types = c("integer", "numeric", "factor", "ordered", "logical"),
+        predict_types = c("response", "prob"),
+        param_set = param_set,
+        properties = c("twoclass", "multiclass"),
+        man = "mlr3extralearners::mlr_learners_classif.rpf",
+        label = "Random Planted Forest"
+      )
+    }
+  ),
+  private = list(
+    .train = function(task) {
+      # get parameters for training
+      pars = self$param_set$get_values(tags = "train")
+      # max_interaction_limit is needed but must not be passed to rpf(),
+      # while convert_ratio automatically removes max_interaction_ratio.
+      max_interaction_limit = pars[["max_interaction_limit"]]
+      pars[["max_interaction_limit"]] = NULL
+      n_features = length(task$feature_names)
+
+      pars = convert_ratio(
+        pars, "max_interaction", "max_interaction_ratio",
+        min(n_features, max_interaction_limit)
+      )
+
+      invoke(
+        randomPlantedForest::rpf,
+        x = task$data(cols = task$feature_names),
+        y = task$data(cols = task$target_names),
+        .args = pars
+      )
+    },
+    .predict = function(task) {
+      pars = self$param_set$get_values(tags = "predict")
+      newdata = ordered_features(task, self)
+
+      if (self$predict_type == "response") {
+        pred = invoke(
+          predict, self$model, new_data = newdata,
+          type = "class", .args = pars
+        )
+        list(response = pred[[".pred_class"]])
+      } else {
+        pred = invoke(
+          predict, self$model, new_data = newdata,
+          type = "prob", .args = pars
+        )
+        # Result will be a df with one column per variable with names '.pred_<level>'
+        # we want the names without ".pred"
+        xn = names(pred)
+        xn[which(xn == paste0(".pred_", task$class_names))] = task$class_names
+        names(pred) = xn
+
+        list(prob = as.matrix(pred))
+      }
+    }
+  )
+)
+
+.extralrns_dict$add("classif.rpf", LearnerClassifRandomPlantedForest)

R/learner_randomPlantedForest_regr_rpf.R

@@ -0,0 +1,90 @@
+#' @title Regression Random Planted Forest Learner
+#' @author jemus42
+#' @name mlr_learners_regr.rpf
+#'
+#' @description
+#' Random Planted Forest: A directly interpretable tree ensemble.
+#'
+#' Calls [randomPlantedForest::rpf()] from 'randomPlantedForest'.
+#'
+#' @inheritSection mlr_learners_classif.rpf Custom mlr3 parameters
+#' @templateVar id regr.rpf
+#' @template learner
+#' @inheritSection mlr_learners_classif.rpf Installation
+#'
+#' @references
+#' `r format_bib("hiabu_2023")`
+#'
+#' @template seealso_learner
+#' @template example
+#' @export
+LearnerRegrRandomPlantedForest = R6Class("LearnerRegrRandomPlantedForest",
+  inherit = LearnerRegr,
+  public = list(
+    #' @description
+    #' Creates a new instance of this [R6][R6::R6Class] class.
+    initialize = function() {
+      param_set = ps(
+        max_interaction       = p_int(lower = 0, upper = Inf, default = 1, tags = "train"),
+        max_interaction_ratio = p_dbl(lower = 0, upper = 1, tags = "train"),
+        max_interaction_limit = p_int(lower = 1, upper = Inf, tags = c("required", "train")),
+        ntrees                = p_int(lower = 1, upper = Inf, default = 50, tags = "train"),
+        splits                = p_int(lower = 1, upper = Inf, default = 30, tags = "train"),
+        split_try             = p_int(lower = 1, upper = Inf, default = 10, tags = "train"),
+        t_try                 = p_dbl(lower = 0, upper = 1, default = 0.4, tags = "train"),
+        deterministic         = p_lgl(default = FALSE, tags = "train"),
+        nthreads              = p_int(lower = 1, upper = Inf, default = 1, tags = c("train", "threads")),
+        cv                    = p_lgl(default = FALSE, tags = "train"),
+        purify                = p_lgl(default = FALSE, tags = "train")
+      )
+
+      param_set$values = list(max_interaction_limit = Inf)
+
+      super$initialize(
+        id = "regr.rpf",
+        packages = "randomPlantedForest",
+        feature_types = c("integer", "numeric", "factor", "ordered", "logical"),
+        predict_types = "response",
+        param_set = param_set,
+        properties = character(0),
+        man = "mlr3extralearners::mlr_learners_regr.rpf",
+        label = "Random Planted Forest"
+      )
+    }
+  ),
+  private = list(
+    .train = function(task) {
+      # get parameters for training
+      pars = self$param_set$get_values(tags = "train")
+      # max_interaction_limit is needed but must not be passed to rpf(),
+      # while convert_ratio automatically removes max_interaction_ratio.
+      max_interaction_limit = pars[["max_interaction_limit"]]
+      pars[["max_interaction_limit"]] = NULL
+      n_features = length(task$feature_names)
+
+      pars = convert_ratio(
+        pars, "max_interaction", "max_interaction_ratio",
+        min(n_features, max_interaction_limit)
+      )
+
+      invoke(
+        randomPlantedForest::rpf,
+        x = task$data(cols = task$feature_names),
+        y = task$data(cols = task$target_names),
+        .args = pars
+      )
+    },
+    .predict = function(task) {
+      pars = self$param_set$get_values(tags = "predict")
+      newdata = ordered_features(task, self)
+
+      pred = invoke(
+        predict, self$model, new_data = newdata,
+        type = "numeric", .args = pars
+      )
+      list(response = pred[[".pred"]])
+    }
+  )
+)
+
+.extralrns_dict$add("regr.rpf", LearnerRegrRandomPlantedForest)