Skip to contents

Fits the boosting algorithm from `ranktreeEnsemble` 0.24 using the same fold-local, collision-free feature mapping and deployable `datrank=FALSE` contract as [fit_reo_rankforest()]. The backend's exposed `weights` argument is not forwarded by version 0.24 and is therefore not represented as active in this adapter.

Usage

fit_reo_rankboost(X_train, y_train, meta_train = NULL, hp = list())

Arguments

X_train

Numeric non-negative matrix of training samples by features.

y_train

Numeric/integer 0/1 training labels; 1 is the case class.

meta_train

Optional training metadata, checked for row alignment and otherwise ignored.

hp

Optional fixed hyperparameters: `ntree` (500), `seed` (1), `dimreduce` (`TRUE`), `nodedepth` (3), `nodesize` (5), `shrinkage` (0.05), requested `bag_fraction` (0.5), and `max_screen_features` (128). The effective bag fraction applies a conservative training-size-only one-observation margin above the pinned `gbm` node-size boundary; both the requested and effective values are retained in the fitted state. A training set with no variable pair ordering returns an exact intercept-only state.

Value

A frozen `reo_rankboost_model` suitable for singleton scoring.