Regression Random Planted Forest Learner

Random planted forest: a directly interpretable tree ensemble. Calls randomPlantedForest::rpf() from package 'randomPlantedForest'.

Dictionary

This Learner can be instantiated via lrn():

lrn("regr.rpf")

Meta Information

• Task type: “regr”

• Predict Types: “response”

• Feature Types: “logical”, “integer”, “numeric”, “factor”, “ordered”

• Required Packages: mlr3, mlr3extralearners, randomPlantedForest

Parameters

Id	Type	Default	Levels	Range
max_interaction	integer	2		\([0, \infty)\)
max_interaction_ratio	numeric	-		\([0, 1]\)
max_interaction_limit	integer	-		\([1, \infty)\)
ntrees	integer	50		\([1, \infty)\)
splits	integer	30		\([1, \infty)\)
split_structure	character	leaves	leaves, hist, cur_trees_1, cur_trees_2, res_trees	-
split_try	integer	10		\([1, \infty)\)
t_try	numeric	0.4		\([0, 1]\)
max_candidates	integer	50		\([1, \infty)\)
split_decay_rate	numeric	0.1		\([0, \infty)\)
delete_leaves	logical	TRUE	TRUE, FALSE	-
purify	logical	FALSE	TRUE, FALSE	-
nthreads	integer	1		\([1, \infty)\)
export_forest	logical	FALSE	TRUE, FALSE	-
deterministic	logical	FALSE	TRUE, FALSE	-
marshal_include_data	logical	FALSE	TRUE, FALSE	-

Installation

Package 'randomPlantedForest' is not on CRAN and has to be installed from GitHub via remotes::install_github("PlantedML/randomPlantedForest").

Saving a Learner

The fitted forest lives in C++ memory and does not survive R serialization. Call $marshal() before writing a trained learner to disk, and $unmarshal() after loading it again. By default, marshaling does not store the training data, so a restored model cannot be purified afterwards. Set purify = TRUE at training time if purification is needed, or include the training data in the marshaled model via the marshal_include_data parameter or $marshal(include_data = TRUE).

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, with max_interaction_limit as an additional constraint. The parameter max_interaction_limit is initialized to Inf.

• marshal_include_data:

  • Whether to store the training data when marshaling the model, which is required to call randomPlantedForest::purify() on a restored model. Defaults to FALSE to keep marshaled models small. Can be overridden for a single call via $marshal(include_data = TRUE).

References

Hiabu, Munir, Mammen, Enno, Meyer, T. J (2023). “Random Planted Forest: a directly interpretable tree ensemble.” arXiv preprint arXiv:2012.14563. doi:10.48550/ARXIV.2012.14563 .

See also

• Dictionary of Learners: mlr3::mlr_learners.

• as.data.table(mlr_learners) for a table of available Learners in the running session (depending on the loaded packages).

• Chapter in the mlr3book: https://mlr3book.mlr-org.com/chapters/chapter2/data_and_basic_modeling.html#sec-learners

• mlr3learners for a selection of recommended learners.

• mlr3cluster for unsupervised clustering learners.

• mlr3pipelines to combine learners with pre- and postprocessing steps.

• mlr3tuning for tuning of hyperparameters, mlr3tuningspaces for established default tuning spaces.

Author

jemus42

Super classes

mlr3::Learner -> mlr3::LearnerRegr -> LearnerRegrRandomPlantedForest

Active bindings

marshaled

(logical(1))
Whether the learner has been marshaled.

Methods

Public methods

• LearnerRegrRandomPlantedForest$new()

• LearnerRegrRandomPlantedForest$marshal()

• LearnerRegrRandomPlantedForest$unmarshal()

• LearnerRegrRandomPlantedForest$clone()

Inherited methods

• mlr3::Learner$base_learner()
• mlr3::Learner$configure()
• mlr3::Learner$encapsulate()
• mlr3::Learner$format()
• mlr3::Learner$help()
• mlr3::Learner$predict()
• mlr3::Learner$predict_newdata()
• mlr3::Learner$print()
• mlr3::Learner$reset()
• mlr3::Learner$selected_features()
• mlr3::Learner$train()
• mlr3::LearnerRegr$predict_newdata_fast()

---

LearnerRegrRandomPlantedForest$new()

Creates a new instance of this R6 class.

Usage

LearnerRegrRandomPlantedForest$new()

---

LearnerRegrRandomPlantedForest$marshal()

Marshal the learner's model.

Usage

LearnerRegrRandomPlantedForest$marshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::marshal_model().

---

LearnerRegrRandomPlantedForest$unmarshal()

Unmarshal the learner's model.

Usage

LearnerRegrRandomPlantedForest$unmarshal(...)

Arguments

...

(any)
Additional arguments passed to mlr3::unmarshal_model().

---

LearnerRegrRandomPlantedForest$clone()

The objects of this class are cloneable with this method.

Usage

LearnerRegrRandomPlantedForest$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner
learner = lrn("regr.rpf")
print(learner)
#> 
#> ── <LearnerRegrRandomPlantedForest> (regr.rpf): Random Planted Forest ──────────
#> • Model: -
#> • Parameters: max_interaction_limit=Inf
#> • Packages: mlr3, mlr3extralearners, and randomPlantedForest
#> • Predict Types: [response]
#> • Feature Types: logical, integer, numeric, factor, and ordered
#> • Encapsulation: none (fallback: -)
#> • Properties: marshal
#> • Other settings: use_weights = 'error', predict_raw = 'FALSE'

# Define a Task
task = tsk("mtcars")

# Create train and test set
ids = partition(task)

# Train the learner on the training ids
learner$train(task, row_ids = ids$train)

print(learner$model)
#> ── Regression Random Planted Forest ────────────────────────────────────────────
#> Predictors: `am`, `carb`, `cyl`, …, `vs`, and `wt`
#> 50 tree families with 30 splits each on 10 predictors, interactions to degree
#> 2.
#> ℹ Forest is not purified.
#> 
#> ── Tree growing 
#>    split_structure: leaves
#>          split_try: 10
#>              t_try: 0.4
#>     max_candidates: 50
#>   split_decay_rate: 0.1
#>      delete_leaves: TRUE
#> 
#> ℹ Fit using 1 thread, also the default for `predict()` and `purify()`.


# Make predictions for the test rows
predictions = learner$predict(task, row_ids = ids$test)

# Score the predictions
predictions$score()
#> regr.mse 
#>  7.91502