Personal ML template

Disclaimer: Even if it is a personal project, everybody can use it freely and modify it for their own needs.

This repository is a template for using in ML projects. It includes:

• Inference and training template scripts
• ⚙️ YAECS as configuration manager (compatible with WandB, ClearML, ...)
• ✅ pytest-cov to check unit tests and get coverage (including an optional minimum coverage to pass)
• 🎨 ruff to check the style and auto-format it, including:
  • pycodestyle and flake to check overall Python scripts style (PEP8)
  • isort to check the import order of Python scripts
  • pydocstyle to check Python docstrings style (Numpy convention)
• 🎨 pylint to have an overall grade of the style (including an optional minimum grade to pass)
• 🎨 black to auto-format Python scripts
• 🏷️ mypy to check typing and type hints

Some GitHub actions 🏭 are provided:

• 🔒 bandit for security
• 🎨 flake8, mypy, pydocstyle, pylint, ruff for style
• Cache is preserved between GH action runs (useful for heavy requirements packages)

And finally, some badges:

• 🆙 Release, last commit and release date
• 🧑‍🤝‍🧑 Github stats
• 📑 licenses
• 🔖 python version (and pytorch version)
• Pytorch and Wandb badges
• And more...

All workflows create a badge available, for instance, in README.

This repository provides also a pre-commit configuration to check end-of-file, trailing whitespace, flake8 and pydocstyle (numpy).

HowTo

All feature of this template is easy to adapt on your project by changing names or versions on the .github/workflows/ directory and on the badge paths on your markdown/rst files. All the worflows are independent and can be used individually. You can also remove any workflow you don't need.

Before all, you need to create a gist The id of the gist is required for pylint and test/coverage badges. Then, you must add a secret in your repository (Settings > Secrets > New repository secret) that is a personal token with gist scope with name GIST_SECRET (details here).

Notes

By default, there is no maximum unit test coverage but you can set the minimum coverage you want in utils/github_actions/pytest_manager.py. There is also a minimum grade for pylint that is 7.0/10 and can be set in utils/github_actions/pylint_manager.py. Details of pylint options are in .pylintrc and can also be changed at will.

Contributing

Even if it is a personal template, feel free to contribute via issues or pull requests 🤗.

See CONTRIBUTING.md for more details.