Example gallery: illustrations and external evidence • sglasso

Grouped coefficient paths

Six Gaussian coefficient paths, colored by three prespecified groups

This is the small Gaussian quick-start example: seed 19, 80 synthetic observations and six predictors in three groups. The horizontal axis is penalty strength on a log scale; the vertical axis is the coefficient value on the original predictor scale. It demonstrates the interface, not comparative performance or true-support recovery.

Independent external-test ROC curves

Five ROC curves from the verified independent CoRSIVSZ test cohort

These curves use the already fitted, fixed predictions for the 675-person external CoRSIVSZ cohort. They are not from a new refit, random train/test split or resampling run. All five curves use the same line width. AUC measures ranking, not accuracy at a selected decision threshold. SGLASSO’s observed score does not by itself establish multiplicity-adjusted superiority. The extra d = 0 audit variant remains in the run code, not this five-method display. Full uncertainty is reported in the manuscript, not inferred from the visual separation of curves.

Existing Gaussian documentation plots

Existing GenAtHum documentation plot of CV error against lambda

Existing GenAtHum documentation plot of CV error against the target scale d

These original package-documentation graphics are retained unchanged. They illustrate the GenAtHum example, not the new binary analysis or a new benchmark. The historical generating example is shown below and is not executed when building the site:

library(sglasso)
data(GenAtHum, package="sglasso")
set.seed(2025)
model_CV <- cv.sglasso(GenAtHum$X, GenAtHum$y, GenAtHum$group,
                       nlambda=20, nd=5, nfolds=5, alpha=0.4)
plot(model_CV)
plot(model_CV, type.tun="d")

Historical figure provenance does not establish bit-for-bit agreement with every future package version. The original article scripts retain their own study settings.

Reproduction code

tools/render_sglasso_gallery.R draws the first figure from the documented tiny example and the ROC figure from a checksum-verified completed external run. It refuses to overwrite existing output files. The ROC rendering step requires the completed external outputs; these are not bundled with the code. It never reruns the external models or bootstrap.