Contributing

This is a personal helpers package. Keep the public surface small: cf.io, cf.data, cf.pl, cf.stats, cf.ext.

Scanpy’s developer documentation covers git, tests, and docstrings. The notes below are cellfish-specific.

Install

cd cellfish
python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev,plot,doc]"

Tests:

pytest

Adding an algorithm

1. Create src/cellfish/ext/<tool>/__init__.py plus one implementation file.

2. Keep compute and plotting together; reuse cf.pl for fonts, palettes, and grids.

3. Put optional dependencies in pyproject.toml extras and import them inside functions.

4. Add a smoke test with a tiny simulated object.

5. Do not add plot/_<tool>.py, and do not change layer 1 unless two tools share the logic.

Layer 1 (io / data / plot / stats) must not import ext.

Documentation

This project uses Sphinx with MyST markdown, Napoleon docstrings, and autosummary — the same layout as scMagnify.

• New public functions need a docstring (numpy or google style) and an entry in docs/api/.

• Tutorials live in docs/tutorials/ and docs/notebooks/.

• Citations use {cite:p}Key`` once docs/references.bib has the entry.

If you refer to objects from other packages, add them to intersphinx_mapping in docs/conf.py. If a missing link is outside your control, add it to nitpick_ignore.

Build locally

pip install -e ".[doc,plot]"
cd docs
make html
python -m webbrowser -t _build/html/index.html

Generated autosummary pages land in docs/api/generated/ and are not committed.

Code style

Optional pre-commit and ruff are listed in pyproject.toml. Do not expand layer 1 to absorb paper-specific palettes or figure panels.