Skip to content

README.md

Latest commit

 

History

History
66 lines (48 loc) · 4.57 KB

README.md

File metadata and controls

66 lines (48 loc) · 4.57 KB
Raw

R-CMD-check lint cyclocomp autotest goodpractice pkgcheck Codecov test coverage life cycle: experimental CRAN status

Sparse modelling with grouped and correlated features

Scope

Implements lasso regression (Gaussian, logistic, Poisson, and Cox) for features in heterogeneous groups (e.g., where some groups contain subgroups of correlated features), overlapping groups (i.e., where some features belong to two or more groups), and misspecified groups (i.e., where some group labels are incorrect or missing). Involves an initial ridge regression followed by a final adaptive lasso regression. Exchanges information between features according to their assumed adjacencies or their observed correlations to improve selection and predictive performance. Allows for features that can provide information on other features but must not themselves be selected (i.e., privileged information).

Installation

Install the latest development version from GitHub:

#install.packages("remotes")
remotes::install_github("rauschenberger/corila")

Usage

Use the function cv.corila to model an outcome (n-dimensional vector y) based on many predictors (n x p matrix X) that are structured by groups (e.g., p-dimensional vector group) and potentially split into primary and auxiliary predictors (p-dimensional vector primary). See the vignette for detailed examples.

library(corila)
data <- simulate_data()
object <- cv.corila(x = data$x_train,
                    y = data$y_train,
                    group = data$group,
                    primary = data$primary)
coef(object)
predict(object, newx = data$x_test)

Reference

Armin Rauschenberger AR (2026). "Sparse modelling with grouped and correlated features allowing for privileged information". Manuscript in preparation.

(Presented at SIS-FENStatS in Rome, Italy, on 25 June 2026.)

Disclaimer

This public repository is on a personal GitHub account, but it has private pull mirrors on two institutional GitLab instances (see https://git.lih.lu/arauschenberger/corila and https://gitlab.com/uniluxembourg/Personalfolders/armin.rauschenberger/corila)).

Large-language models (mainly Claude Sonnet 4.6 and GPT-5.4) were used for reviewing R code and documentation and for adapting configuration files (.yaml).

Copyright © 2025 Armin Rauschenberger; Luxembourg Institute of Health (LIH), Department of Medical Informatics (DMI), Bioinformatics and Artificial Intelligence (BioAI); University of Luxembourg, Luxembourg Centre for Systems Biomedicine (LCSB), Biomedical Data Science (BDS). This R package is distributed under a GPL 3 license.