robustStepwiseSplit: Robust Multi-Model Stepwise Variable Selection

In robStepSplitReg: Robust Stepwise Split Regularized Regression

Robust Multi-Model Stepwise Variable Selection

Description

Internal function that performs robust stepwise variable selection for building disjoint ensemble models. Uses robust correlation estimates to select predictor subsets while ensuring variables are not shared across models.

Usage

robustStepwiseSplit(
  Rx,
  Ry,
  n_models = 1,
  model_saturation = "p-value",
  alpha = 0.05,
  model_size = NULL,
  n = nrow(Rx)
)

Arguments

Rx	Robust correlation matrix of predictors (p x p matrix).
Ry	Robust correlation vector between response and predictors (p x 1 vector).
n_models	Number of models to build. Default is 1.
model_saturation	Criterion to determine model saturation. Either "p-value" or "fixed". Default is "p-value".
alpha	P-value threshold for determining model saturation when model_saturation = "p-value". Default is 0.05.
model_size	Maximum number of variables per model when model_saturation = "fixed". Default is NULL.
n	Sample size used for computing test statistics. Default is nrow(Rx).

Value

If n_models = 1, returns a vector of 0-indexed selected variable indices. If n_models > 1, returns a list where each element contains the 0-indexed selected variable indices for the corresponding model.

Author(s)

Anthony-Alexander Christidis, [e-mail]

robStepSplitReg documentation built on Aug. 21, 2025, 5:26 p.m.