Update {regr,classif}.ranger learner for 0.17.0 (#321)

* refactor: update {regr,classif}.ranger learner for 0.17.0

* chore: update NEWS.md

* chore: fix paramtest fro classif.ranger to exlucde regr-only param

* docs: add new ranger num.threads behavior docs

* chore: roxygenize

* refactor: make min.bucket and min.node.size a p_uty for classif.ranger to allow vector values

NEWS.md

@@ -1,6 +1,8 @@
 # mlr3learners (development version)
 
-compatibility: mlr3 0.22.0.
+* feat: update hyperparameter set of `lrn("classif.ranger")` and `lrn("regr.ranger")` for 0.17.0, adding `na.action` parameter and `"missings"` property, and `poisson` splitrule for regression with a new `poisson.tau` parameter.
+* compatibility: mlr3 0.22.0.
+
 
 # mlr3learners 0.8.0
 

R/LearnerClassifRanger.R

@@ -14,7 +14,9 @@
 #'
 #' @section Initial parameter values:
 #' - `num.threads`:
-#'   - Actual default: `NULL`, triggering auto-detection of the number of CPUs.
+#'   - Actual default: `2`, using two threads, while also respecting environment variable
+#'     `R_RANGER_NUM_THREADS`, `options(ranger.num.threads = N)`, or `options(Ncpus = N)`, with
+#'     precedence in that order.
 #'   - Adjusted value: 1.
 #'   - Reason for change: Conflicting with parallelization via \CRANpkg{future}.
 #'
@@ -42,10 +44,19 @@ LearnerClassifRanger = R6Class("LearnerClassifRanger",
         importance                   = p_fct(c("none", "impurity", "impurity_corrected", "permutation"), tags = "train"),
         keep.inbag                   = p_lgl(default = FALSE, tags = "train"),
         max.depth                    = p_int(default = NULL, lower = 1L, special_vals = list(NULL), tags = "train"),
-        min.bucket                   = p_int(1L, default = 1L, tags = "train"),
-        min.node.size                = p_int(1L, default = NULL, special_vals = list(NULL), tags = "train"),
+        min.bucket                   = p_uty(default = 1L, tags = "train",
+                                             custom_check = function(x) {
+                                               if (checkmate::test_integerish(x)) return(TRUE)
+                                               "Must be integer of length 1 or greater"
+                                             }),
+        min.node.size                = p_uty(default = NULL, special_vals = list(NULL), tags = "train",
+                                             custom_check = function(x) {
+                                               if (checkmate::test_integerish(x, null.ok = TRUE)) return(TRUE)
+                                               "Must be integer of length 1 or greater"
+                                             }),
         mtry                         = p_int(lower = 1L, special_vals = list(NULL), tags = "train"),
         mtry.ratio                   = p_dbl(lower = 0, upper = 1, tags = "train"),
+        na.action                    = p_fct(c("na.learn", "na.omit", "na.fail"), default = "na.learn", tags = "train"),
         num.random.splits            = p_int(1L, default = 1L, tags = "train", depends = quote(splitrule == "extratrees")),
         node.stats                   = p_lgl(default = FALSE, tags = "train"),
         num.threads                  = p_int(1L, default = 1L, tags = c("train", "predict", "threads")),
@@ -72,7 +83,7 @@ LearnerClassifRanger = R6Class("LearnerClassifRanger",
         param_set = ps,
         predict_types = c("response", "prob"),
         feature_types = c("logical", "integer", "numeric", "character", "factor", "ordered"),
-        properties = c("weights", "twoclass", "multiclass", "importance", "oob_error", "hotstart_backward"),
+        properties = c("weights", "twoclass", "multiclass", "importance", "oob_error", "hotstart_backward", "missings"),
         packages = c("mlr3learners", "ranger"),
         label = "Random Forest",
         man = "mlr3learners::mlr_learners_classif.ranger"

R/LearnerRegrRanger.R

@@ -36,11 +36,13 @@ LearnerRegrRanger = R6Class("LearnerRegrRanger",
         min.node.size                = p_int(1L, default = 5L, special_vals = list(NULL), tags = "train"),
         mtry                         = p_int(lower = 1L, special_vals = list(NULL), tags = "train"),
         mtry.ratio                   = p_dbl(lower = 0, upper = 1, tags = "train"),
+        na.action                    = p_fct(c("na.learn", "na.omit", "na.fail"), default = "na.learn", tags = "train"),
         node.stats                   = p_lgl(default = FALSE, tags = "train"),
         num.random.splits            = p_int(1L, default = 1L, tags = "train", depends = quote(splitrule == "extratrees")),
         num.threads                  = p_int(1L, default = 1L, tags = c("train", "predict", "threads")),
         num.trees                    = p_int(1L, default = 500L, tags = c("train", "predict", "hotstart")),
         oob.error                    = p_lgl(default = TRUE, tags = "train"),
+        poisson.tau                  = p_dbl(default = 1, tags = "train", depends = quote(splitrule == "poisson")),
         regularization.factor        = p_uty(default = 1, tags = "train"),
         regularization.usedepth      = p_lgl(default = FALSE, tags = "train"),
         replace                      = p_lgl(default = TRUE, tags = "train"),
@@ -51,7 +53,7 @@ LearnerRegrRanger = R6Class("LearnerRegrRanger",
         se.method                    = p_fct(c("jack", "infjack"), default = "infjack", tags = "predict"), # FIXME: only works if predict_type == "se". How to set dependency?
         seed                         = p_int(default = NULL, special_vals = list(NULL), tags = c("train", "predict")),
         split.select.weights         = p_uty(default = NULL, tags = "train"),
-        splitrule                    = p_fct(c("variance", "extratrees", "maxstat"), default = "variance", tags = "train"),
+        splitrule                    = p_fct(c("variance", "extratrees", "maxstat", "beta", "poisson"), default = "variance", tags = "train"),
         verbose                      = p_lgl(default = TRUE, tags = c("train", "predict")),
         write.forest                 = p_lgl(default = TRUE, tags = "train")
       )
@@ -63,7 +65,7 @@ LearnerRegrRanger = R6Class("LearnerRegrRanger",
         param_set = ps,
         predict_types = c("response", "se", "quantiles"),
         feature_types = c("logical", "integer", "numeric", "character", "factor", "ordered"),
-        properties = c("weights", "importance", "oob_error", "hotstart_backward"),
+        properties = c("weights", "importance", "oob_error", "hotstart_backward", "missings"),
         packages = c("mlr3learners", "ranger"),
         label = "Random Forest",
         man = "mlr3learners::mlr_learners_regr.ranger"

inst/paramtest/test_paramtest_classif.ranger.R

@@ -14,6 +14,7 @@ test_that("classif.ranger", {
     "class.weights", # handled via mlr3
     "inbag", # handled via mlr3 stratification
     "quantreg", # regression only
+    "poisson.tau", # regression only
     "dependent.variable.name", # handled via mlr3
     "status.variable.name", # handled via mlr3
     "classification", # handled via mlr3