Install

Set up R and Python to follow along live

If you would like to run the demos on your laptop, install one or both of the combss packages below. The same demos exist in R and Python with matching results, so either language is fine.

R

install.packages("combss")

That’s it — combss is on CRAN. The only run-time dependency is glmnet, which CRAN will install automatically. Tested with R 4.3 and 4.4 on macOS, Linux, and Windows.

Quick check

library(combss)
set.seed(2026)
x <- matrix(rnorm(100 * 20), 100, 20)
y <- as.numeric(x %*% c(2, -1, 1, rep(0, 17)) + rnorm(100) * 0.3)
fit <- combss(x, y, family = "gaussian", q = 5)
fit$subset_list[1:5]

Expected output: nested supports building up to {1, 2, 3} at $k = 3$.

Python

pip install combss

Python 3.9 or newer. Dependencies (numpy, scipy, scikit-learn, pandas) install automatically.

Quick check

import numpy as np
from combss import linear

rng = np.random.default_rng(1)
X = rng.standard_normal((100, 20))
beta = np.r_[2, -1, 1, np.zeros(17)]
y = X @ beta + 0.3 * rng.standard_normal(100)

m = linear.model()
m.fit(X, y, q=5, verbose=False)
print(m.subset_list[:3])

Expected output: nested supports building up to [0, 1, 2] at $k = 3$ (Python is 0-indexed).

Editor recommendations

Either of these makes running the demos chunk-by-chunk smooth:

RStudio — open the .qmd file, click the green ▶ on each chunk.
VS Code with the Quarto and Jupyter extensions — same idea, language-aware.

Cloning the demo repository

git clone https://github.com/saratmoka/combss-statfest.git
cd combss-statfest

The Khan SRBCT train/test CSVs live under data/; simulations use inline set.seed() / np.random.default_rng() calls. The rice GWAS demo is precomputed (the raw SNP matrix is too large to ship), and the figures are in figures/rice/. No external downloads at run time.

Print-friendly handouts (conda from scratch)

If you’d rather start completely fresh — clean conda environment, paste-and-run code, no Quarto — these two PDFs walk you through the same demos as the site, end-to-end:

handout-r.pdf — R quick-start (conda env, R inside the env, RStudio Desktop, Demos 1–4)
handout-python.pdf — Python quick-start (same env, Spyder, Demos 1–4)

Trouble?

R cannot find the package: make sure you are on R ≥ 4.0 and check available.packages() lists combss. If not, your CRAN mirror may be stale; pick another with chooseCRANmirror().
Python pip install fails on combss: confirm you are on Python ≥ 3.9 with python --version; older versions are unsupported.
Anything else: open an issue at https://github.com/saratmoka/combss-statfest/issues.

--- title: "Install" subtitle: "Set up R and Python to follow along live" --- If you would like to run the demos on your laptop, install one or both of the `combss` packages below. The same demos exist in R and Python with matching results, so either language is fine. ## R ```r install.packages("combss") ``` That's it — `combss` is on CRAN. The only run-time dependency is `glmnet`, which CRAN will install automatically. Tested with R 4.3 and 4.4 on macOS, Linux, and Windows. ### Quick check ```r library(combss) set.seed(2026) x <- matrix(rnorm(100 * 20), 100, 20) y <- as.numeric(x %*% c(2, -1, 1, rep(0, 17)) + rnorm(100) * 0.3) fit <- combss(x, y, family = "gaussian", q = 5) fit$subset_list[1:5] ``` Expected output: nested supports building up to `{1, 2, 3}` at $k = 3$. ## Python ```bash pip install combss ``` Python 3.9 or newer. Dependencies (`numpy`, `scipy`, `scikit-learn`, `pandas`) install automatically. ### Quick check ```python import numpy as np from combss import linear rng = np.random.default_rng(1) X = rng.standard_normal((100, 20)) beta = np.r_[2, -1, 1, np.zeros(17)] y = X @ beta + 0.3 * rng.standard_normal(100) m = linear.model() m.fit(X, y, q=5, verbose=False) print(m.subset_list[:3]) ``` Expected output: nested supports building up to `[0, 1, 2]` at $k = 3$ (Python is 0-indexed). ## Editor recommendations Either of these makes running the demos chunk-by-chunk smooth: - **RStudio** — open the `.qmd` file, click the green ▶ on each chunk. - **VS Code** with the *Quarto* and *Jupyter* extensions — same idea, language-aware. ## Cloning the demo repository ```bash git clone https://github.com/saratmoka/combss-statfest.git cd combss-statfest ``` The Khan SRBCT train/test CSVs live under `data/`; simulations use inline `set.seed()` / `np.random.default_rng()` calls. The rice GWAS demo is precomputed (the raw SNP matrix is too large to ship), and the figures are in `figures/rice/`. No external downloads at run time. ## Print-friendly handouts (conda from scratch) If you'd rather start completely fresh — clean `conda` environment, paste-and-run code, no Quarto — these two PDFs walk you through the same demos as the site, end-to-end: - [`handout-r.pdf`](handouts/handout-r.pdf){target="_blank"} — R quick-start (conda env, R inside the env, RStudio Desktop, Demos 1–4) - [`handout-python.pdf`](handouts/handout-python.pdf){target="_blank"} — Python quick-start (same env, Spyder, Demos 1–4) ## Trouble? - R cannot find the package: make sure you are on R ≥ 4.0 and check `available.packages()` lists `combss`. If not, your CRAN mirror may be stale; pick another with `chooseCRANmirror()`. - Python `pip install` fails on `combss`: confirm you are on Python ≥ 3.9 with `python --version`; older versions are unsupported. - Anything else: open an issue at <https://github.com/saratmoka/combss-statfest/issues>.