Cookiecutter Data Science

Cookiecutter Data Science (CCDS) is a free, open-source template that gives Python data science and machine learning projects a standardized, reproducible directory layout. Instead of every analyst inventing their own folder scheme, you get consistent homes for raw, interim, processed, and external data, plus src code, notebooks, models, reports, figures, references, and docs — so a teammate can open your repo and know where things live. CCDS v2 ships the `ccds` CLI, which turns project setup into an interactive wizard. It walks you through environment manager (virtualenv, conda, pipenv, uv, pixi, poetry, or none), dependency file (requirements.txt, pyproject.toml, environment.yml, Pipfile, pixi.toml), testing framework (pytest, unittest, none), linting and formatting (ruff or flake8+black+isort), docs generator (mkdocs or none), dataset storage (Azure, AWS S3, GCS, or none), Python version, and an open-source license. Install via pipx, pip, or - per the docs, coming soon - conda, then run `ccds` from the parent directory and answer the prompts. What lands on disk is opinionated but transparent: a Makefile with convenience targets like `make data` and `make train`, a numbered Jupyter notebook naming convention (e.g. `1.0-jqp-initial-data-exploration`), an optional source-code scaffold that turns your module into an importable Python package with config, dataset, features, modeling/train, modeling/predict, and plots scripts, and a mkdocs project when you want it. You can also point `ccds` at your own template URL instead of the default. It suits data scientists, analysts, and small teams who want a lightweight, customizable starting point rather than a full platform. CCDS deliberately provides structure, not orchestration — you bring pipeline and versioning tools like DVC or Kedro. Compared with heavyweight MLOps suites or hand-rolled folder conventions, CCDS stays free, readable, and easy to strip down.