combss-statfest

Companion repository for the StatFest talk COMBSS: Scalable Best Subset Selection for Generalised Linear Models (Sarat Moka, UNSW Sydney).

The repo contains runnable simulation and real-data demos in R and Python for the combss packages. Slides for the talk itself are produced separately in Keynote.

Structure

combss-statfest/
├── _quarto.yml            # site config (HTML + PDF)
├── index.qmd              # landing
├── install.qmd            # R + Python install steps
├── demos/
│   ├── 01-simulation.qmd  # linear sim, n=300, p=30
│   ├── 02-khan.qmd        # Khan SRBCT, n=83, p=2308
│   └── 03-comparisons.qmd # head-to-head vs lasso
├── data/
│   ├── Khan_train.csv
│   └── Khan_test.csv
├── figures/               # static images (optional)
└── styles.css

Build

# preview locally (live reload)
quarto preview

# one-shot render (HTML)
quarto render

# render the printable handout
quarto render --to pdf

The HTML site lands in _site/; the PDFs sit alongside each .qmd after a PDF render.

Run the demos yourself

1. Install the packages — see install.qmd.
2. Open a .qmd in RStudio or VSCode and run the chunks interactively, or just read the rendered HTML.

Data

• data/Khan_train.csv and data/Khan_test.csv — Khan et al. (2001) SRBCT microarray, $n_{\text{train}} = 63$, $n_{\text{test}} = 20$, $p = 2308$ genes, 4 tumour classes.
• Other demos are seeded simulations — no external data needed.

Packages

• R — combss on CRAN
• Python — combss on PyPI