Guided Regularized Random Forest Learner

An mlr3::LearnerClassif that fits a Guided Regularized Random Forest (GRRF) for use in mlr3filters::FilterImportance-based feature selection.

The learner implements a two-stage approach:

1. An initial unregularized random forest estimates variable importance scores.

2. Those scores are normalized and used to compute per-feature split-selection weights (coefReg), penalizing less informative predictors.

3. A second "guided regularized" forest is trained with those weights, and its impurity importance scores are exposed via $importance().

Class-balanced sample weights are computed automatically from the target column, so no external weight vector is required.

Parameters

• gamma (numeric [0, 1], default 0.5): Guidance coefficient. 0 = unguided regularized forest (equal penalty for all features); 1 = strongest guiding effect (most important features penalized least).

• num.trees (integer >= 1, default 500): Number of trees in each forest.

• max.depth (integer >= 0, default 100): Maximum tree depth (0 = unlimited).

References

Deng, H., & Runger, G. (2013). Gene selection with guided regularized random forest. Pattern Recognition, 46(12), 3483-3489. https://doi.org/10.1016/j.patcog.2013.05.018 and Wundervald, B. et al. (2020). Generalizing Gain Penalization for Feature Selection in Tree-Based Models. IEEE Access, Vol. 8, 190231 - 190239. https://doi.org/10.1109/ACCESS.2020.3032095

Super classes

mlr3::Learner -> mlr3::LearnerClassif -> LearnerClassifGrrf

Methods

Public methods

• LearnerClassifGrrf$new()

• LearnerClassifGrrf$importance()

• LearnerClassifGrrf$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$train()
• mlr3::LearnerClassif$predict_newdata_fast()

---

LearnerClassifGrrf$new()

Initialise the learner with its parameter set.

Usage

LearnerClassifGrrf$new()

---

LearnerClassifGrrf$importance()

Return impurity importance scores from the fitted GRRF.

Usage

LearnerClassifGrrf$importance()

Returns

Named numeric vector of importance scores, sorted decreasingly. All features present in the training task are included (unused features receive a score of 0).

---

LearnerClassifGrrf$clone()

The objects of this class are cloneable with this method.

Usage

LearnerClassifGrrf$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

if (FALSE) { # \dontrun{
library(mlr3)
library(mlr3filters)

learner <- LearnerClassifGrrf$new()
learner$param_set$values <- list(gamma = 0.9, num.trees = 50L)

filter <- mlr3filters::FilterImportance$new(learner = learner)
task <- mlr3::tsk("sonar")
filter$calculate(task)
as.data.table(filter)
} # }