Merge 4a6d189 into d80e0d1

DESCRIPTION

@@ -49,14 +49,14 @@ URL: https://github.com/mlr-org/mlr3tuning
 BugReports: https://github.com/mlr-org/mlr3tuning/issues
 RoxygenNote: 6.1.1
 Collate:
+    'AutoTuner.R'
     'FitnessFunction.R'
     'Terminator.R'
     'TerminatorEvaluations.R'
     'TerminatorIterations.R'
     'TerminatorMultiplexer.R'
     'Tuner.R'
     'TunerGridSearch.R'
-    'TunerNestedResampling.R'
     'TunerRandomSearch.R'
     'tune_control.R'
     'zzz.R'

NAMESPACE

@@ -1,13 +1,13 @@
 # Generated by roxygen2: do not edit by hand
 
+export(AutoTuner)
 export(FitnessFunction)
 export(Terminator)
 export(TerminatorEvaluations)
 export(TerminatorIterations)
 export(TerminatorMultiplexer)
 export(Tuner)
 export(TunerGridSearch)
-export(TunerNestedResampling)
 export(TunerRandomSearch)
 export(tune_control)
 import(checkmate)

R/AutoTuner.R

@@ -0,0 +1,134 @@
+#' @title AutoTuner
+#'
+#' @description
+#' The [AutoTuner] conducts tuning and sets the learners parameter configuration to the best parameters obtained by the tuning. 
+#' 
+#' Additionally, this class can be used to do nested resampling by passing an [AutoTuner] object to resample.
+#'
+#' @section Usage:
+#' ```
+#' # Construction
+#' at = AutoTuner$new(learner, resampling, param_set, terminator, 
+#'   tuner, tuner_settings, ctrl = tune_control(), id = "autotuner")
+#' 
+#' # public fields
+#' at$learner
+#' ```
+#' See [Learner] for a description of the interface.
+#'
+#' @section Arguments:
+#' * `learner` ([Learner]): \cr
+#'   Internal learner that is tuned and finally returned as trained learner with optimal parameter configuration.
+#' * `resampling` ([Resampling]): \cr
+#'   Resampling strategy for the tuning.
+#' * `param_set` ([paradox::ParamSet]) \cr
+#'   Parameter set for the tuning.
+#' * `terminator` ([Terminator]) \cr
+#'   Terminator used to stop the tuning.
+#' * `tuner` (Tuner Class Generator) \cr
+#'   Uninitialized tuner factory, e.g. TunerGridSearch.
+#' * `tuner_settings` (named list) \cr
+#'   List with tuner settings (e.g. see `?TunerGridSearch`)
+#'
+#' @section Details:
+#' * With `at$learner` the raw learner with the best parameter configuration can be accessed. 
+#'
+#' The interface is described in [Learner].
+#' 
+#' @name AutoTuner
+#' @keywords internal
+#' @family Learner
+#' @examples
+#' task = mlr3::mlr_tasks$get("iris")
+#' learner = mlr3::mlr_learners$get("classif.rpart")
+#' resampling = mlr3::mlr_resamplings$get("holdout")
+#' measures = mlr3::mlr_measures$mget("mmce")
+#' task$measures = measures
+#' param_set = paradox::ParamSet$new(
+#'   params = list(paradox::ParamDbl$new("cp", lower = 0.001, upper = 0.1)))
+#' 
+#' terminator = TerminatorEvaluations$new(5)
+#' 
+#' at = AutoTuner$new(learner, resampling, param_set, terminator, tuner = TunerGridSearch, 
+#'   tuner_settings = list(resolution = 10L))
+#' 
+#' at$train(task)
+#' at$learner
+NULL
+
+#' @export
+AutoTuner = R6Class("AutoTuner", inherit = mlr3::Learner,
+  public = list(
+    learner = NULL,
+
+    initialize = function (learner, resampling, param_set, terminator, tuner, tuner_settings, ctrl = tune_control(), id = "autotuner") {
+      self$learner = mlr3::assert_learner(learner = learner)
+
+      private$.terminator = checkmate::assert_r6(terminator, "Terminator")
+      private$.tuner_settings = checkmate::assert_list(tuner_settings)
+      private$.ff_args$resampling = mlr3::assert_resampling(resampling)
+      private$.ff_args$param_set = checkmate::assert_class(param_set, "ParamSet")
+      private$.ff_args$ctrl = checkmate::assert_list(ctrl, names = "unique")
+
+      # TODO: Check for factory
+      if (! inherits(tuner, "R6ClassGenerator") && grepl(pattern = "Tuner", x = tuner$classname))
+        mlr3misc::stopf("Tuner must be a R6 class generator that creates tuner (e.g. TunerGridSearch).")
+      private$.tuner = tuner
+
+      super$initialize(
+        id = id,
+        task_type = self$learner$task_type,
+        packages = self$learner$packages,
+        feature_types = self$learner$feature_types,
+        predict_types = self$learner$predict_types,
+        param_set = self$learner$param_set,
+        properties = self$learner$properties
+      )
+    },
+
+    train = function (task) {
+      if (private$.is_trained) {
+        logger::log_warn("Learner is already trained.", namespace = "mlr3")
+      } else {
+        task = mlr3::assert_task(task)
+        self$learner = mlr3::assert_learner(learner = self$learner, task = task)
+
+        private$.tuner_settings$terminator = private$.terminator$clone()
+        private$.tuner_settings$ff = FitnessFunction$new(task = task, learner = self$learner, resampling = private$.ff_args$resampling,
+          param_set = private$.ff_args$param_set, ctrl = private$.ff_args$ctrl)
+
+        private$.tuner = do.call(private$.tuner$new, private$.tuner_settings)
+        private$.tuner$tune()
+
+        self$learner$param_vals = private$.tuner$tune_result()$param_vals
+        self$learner$train(task)
+
+        private$.is_trained = TRUE
+
+        return (invisible(self))
+      }
+    },
+
+    predict = function (task) {
+      return (self$learner$predict(task))
+    }
+  ),
+
+  private = list(
+    .ff_args = NULL,
+    .terminator = NULL,
+    .tuner = NULL,
+    .tuner_settings = NULL,
+    .is_trained = FALSE
+  ),
+
+  active = list(
+    tuner = function () {
+      if (private$.is_trained) {
+        return (private$.tuner)
+      } else {
+        return (NULL)
+      }
+    }
+  )
+)

R/FitnessFunction.R

@@ -102,8 +102,8 @@ FitnessFunction = R6Class("FitnessFunction",
       self$task = mlr3::assert_task(task)
       self$learner = mlr3::assert_learner(learner, task = task)
       self$resampling = mlr3::assert_resampling(resampling)
-      self$param_set = assert_class(param_set, "ParamSet")
-      self$ctrl = assert_list(ctrl, names = "unique")
+      self$param_set = checkmate::assert_class(param_set, "ParamSet")
+      self$ctrl = checkmate::assert_list(ctrl, names = "unique")
       self$hooks = list(update_start = list(), update_end = list())
     },