Ball-Codifference Sure Independence Screening

Computes Ball-codifference scores and performs sure independence screening for high-dimensional predictors. The score combines random-ball empirical probabilities with a bounded local codifference weight based on trigonometric characteristic-function contrasts. The functions are intended for marginal screening when moment-based dependence summaries may be unstable or undefined, such as in heavy-tailed data. The implemented methodology is described in Rezapour and Maroufy (2026) "Ball-Codifference Screening for Heavy-Tailed Predictors" .

BCodifSIS

BCodifSIS computes Ball-codifference scores and performs sure independence screening for high-dimensional predictors. It uses compiled C code for the score calculation and R wrappers for screening.

Comparisons with other screening methods should be done outside this package, for example in a separate simulation script.

Reference

The implemented methodology is described in Rezapour and Maroufy (2026), "Ball-Codifference Screening for Heavy-Tailed Predictors", https://arxiv.org/abs/2607.20821.

Installation from source

install.packages("BCodifSIS_0.3.2.tar.gz", repos = NULL, type = "source")

On Windows, source installation requires Rtools matching your R version.

Quick example

library(BCodifSIS)
grid <- seq_len(80)
X <- vapply(seq_len(20), function(j) {
  sin(grid / (j + 1)) + cos(grid * j / 9)
}, numeric(80))
y <- X[, 1] + 0.5 * sin(grid / 3)

fit <- bcodif_sis(X, y, d = 5)
fit$ix
head(fit$complete.info)

quick_check()