Modern survival analysis for Python and R: R-validated, one response model across both languages.

---

Greenwood is a survival analysis library for the study of time-to-event outcomes. This repository holds both language implementations:

Package	Directory	Status
Python	python/	Feature-rich. See the Python README and the documentation site.
R	r/	Early. Surv() and kaplan_meier(), held to the same contracts as Python. See the R README.

Shared across languages

Both packages are native implementations, validated against R's survival package. They replay the same cases, so they agree on inputs and results.

• spec/ holds the language-neutral contracts (Surv(), event_time(), formulas, Kaplan-Meier) and their conformance cases as JSON.
• fixtures/r/ holds numeric fixtures exported from R for the parity tests.
• scripts/ holds the R scripts that generate both.

The R package replays the Surv() conformance cases and the Kaplan-Meier fixtures so far. The Python package replays all of them.

Shared functions keep the same names in both languages (Surv(), metrics). Estimators follow each language's idiom: classes in Python (KaplanMeier) and lower snake case functions returning S3 objects in R (kaplan_meier()). Language-level conventions follow each language (for example, 0-based error locations and NaN/None in Python), and any differences are recorded as adaptations in spec/.

Development

From the repository root:

make install    # Python package (dev extras, into .venv) and R dependencies
make test       # both test suites (make py-test, make r-test)
make check      # both gates: ruff, pyright, pytest, and R CMD check
make py-check   # one package only (or make r-check)

Run make help for every target, and see CONTRIBUTING.md.

License

MIT (c) Richard Iannone. See the LICENSE file.