Stellars Fork - Opinionated CCDS with Copier Support and Enhanced Environment Management

[stellarshenson]

Hi all,

I've been maintaining a fork of CCDS that reflects opinions formed over years of data science and ML engineering work. Sharing here in case it's useful to others with similar workflows.

Key differences from upstream:

• Copier support - template works with both cookiecutter and copier, enabling copier update for existing projects
• Conda local environments - .venv inside project directory instead of global conda envs (configurable)
• Unified dev dependencies - separate requirements-dev.txt or pyproject.toml [dev] extras
• Jupyter kernel auto-registration - via nb_conda_kernels/nb_venv_kernels with ipykernel fallback
• Docker support - optional Dockerfile with uv or pip, Makefile targets for build/run/push
• Environment encryption - .env.enc for committing encrypted secrets
• Model sync targets - sync_models_up/down mirroring data sync pattern
• and more...

Links:

• Repository: https://github.com/stellarshenson/cookiecutter-data-science
• Philosophy doc: https://github.com/stellarshenson/cookiecutter-data-science/blob/master/docs/docs/stellars-philosophy.md

The README has a detailed comparison table. Happy to discuss design choices if anyone's curious.

As per @pjbull comment upon my pull request, this is the place to share with you folks

[stellarshenson]

Happy to have a chat and a presentation if you'd be willing. This fork is based on my and my clients experiences