Open to Everyone

What Makes a Good PyHealth PR

Anyone can contribute a model, dataset, task, or fix to PyHealth — no need to join the Research Initiative. This is a quick-start guide to making your pull request complete and review-ready. For the full walkthrough — dev setup, data access, and per-type file checklists — see the documentation.

Full Contribution Docs View Slide Deck

Learn From Example PRs

The best way to understand our expectations is to study contributions that got it right. Both include complete implementations, synthetic-data tests, docs, and clear PR descriptions.

ChestX-ray14 Dataset + Classification Tasks

An exemplary dataset contribution: full implementation, binary and multilabel tasks, synthetic-data tests, API docs, and example notebooks — with a clear PR description and file guide.

View PR #392

Model Compatibility Updates (PyHealth 2.0)

A great example of maintaining the library: systematic updates across multiple models, backward compatibility where possible, clear documentation of breaking changes, and updated examples and tests.

View PR #610

How PRs Are Evaluated

Every contribution is reviewed against three criteria.

Completeness

Production-ready code: synthetic-data tests, Google-style docstrings, proper base-class inheritance, and working examples.

Relevance to Healthcare

Addresses a healthcare-specific problem or dataset. Grounding in a peer-reviewed publication strengthens the case.

Relevance to the Package

Adds functionality others will reuse, integrates with existing components, and follows established patterns — growing PyHealth thoughtfully.

Quick Checklist

Before opening your PR, make sure you have:

Implementation file inheriting from the right base class (dataset / task / model), with type hints and Google-style docstrings.
Documentation — a new .rst file under docs/api/… and an update to the matching index so it appears in the docs.
Fast tests in tests/core/ using small synthetic data (2–5 patients, no real datasets, no network) that run in milliseconds.
PR description covering who you are (NetID if UIUC), contribution type, a high-level description, and a file guide.
Rebased on the latest sunlabuiuc main, targeting main, with maintainer edits enabled.

Want the full walkthrough?

Dev environment setup, MIMIC data access, code style, and detailed per-type file checklists live in the documentation.

Read the Full Guide