Skip to contents

Changelog

Source: NEWS.md

sglasso 1.2.0 — development source

This describes GitHub development source, not a tagged or CRAN release.

Binomial interface

  • sglasso(..., family = "binomial") accepts numeric/logical 0/1 responses.
  • Training-only groupwise Firth targets and the accelerated RcppArmadillo hybrid solver support target-directed quadratic shrinkage.
  • Binomial CV recomputes preprocessing and targets inside each training fold, selects by log-loss and keeps alpha fixed within a call.
  • Response predictions are probabilities of the outcome coded as one.
  • Finite-grid endpoint warnings, required-target checks and numerical diagnostics are documented. No universal positive-target null-model lambda is claimed.
  • Gaussian numerical routines are unchanged by the binomial integration.

Data helpers

  • CoRSIVSZ documentation describes independent development and external methylation cohorts and annotation-defined predictor groups.
  • Explicit loaders/download helpers verify the separately distributed file against its recorded size and SHA-256. Nothing is downloaded automatically.
  • Public hosting of the release asset remains pending.

Documentation and research code

  • SVG identity, revised README and a pkgdown documentation website.
  • Small executable guides, a data dictionary, gallery and reproduction guide distinguish package usage from article-specific runs.
  • CITATION.cff supplies software metadata and the published Gaussian article as the preferred citation; no DOI is assigned to the unpublished logistic study.
  • The published article citation is completed consistently across documentation: Yüzbaşı, B. and Cao, J. (2026). Collinear Groupwise Selection via Scaled Group Lasso. The American Statistician, 1–23. Advance online publication, 17 September 2026. doi:10.1080/00031305.2026.2709494.
  • Logistic run code is separate from paper_codes/, which belongs to the article in The American Statistician, and contains code/settings only.
  • Issue templates request minimal, non-sensitive reproductions rather than data.
  • No new production fits, CV study or bootstrap study were run for this documentation; only small documentation examples are executed.