machinelearning147/somepackage

Show how to structure a Python project.

Somepackage

Show how to structure a Python project.

Inspired by https://github.com/kennethreitz/samplemod.

Great resources

• https://manikos.github.io/a-tour-on-python-packaging
• https://docs.python-guide.org/writing/structure/
• http://veekaybee.github.io/2017/09/26/python-packaging/
• https://github.com/audreyr/cookiecutter-pypackage

Recommendations

Python 2 or 3?

• Develop your code under Python 3, test it for both Python 2 and Python 3 but consider Python 3 to be the default today.

Split your code into packages, modules, and functions

• All code should be inside some function (except perhaps if __name__ == '__main__':).
• Split long functions into smaller functions.
• If you need to scroll through a function over several screens, it is probably too long.
• Functions should do one thing and one thing only.
• Hide internals with underscores.
• Organize related functions into modules.
• If modules grow too large, split them.
• Import from other modules under somepackage/ using from .somemodule import something.
• Do file I/O on the "outside" of your code, not deep inside.

Classes vs. functions

• Do not overuse classes.
• Prefer immutable data structures.
• Prefer pure functions.

Naming

• Give the subdirectory the same name as your package.
• Before you name your package, check that the name is not taken on https://pypi.org (you may want to upload your package to PyPI one day).

Interfaces

• In somepackage/__init__.py define what should be visible to the outside.
• Use https://semver.org.

Testing

• Test all non-trivial code. I recommend to use https://pytest.org.
• Use Travis CI: https://docs.travis-ci.com/user/languages/python/.

Dependency management

• Package dependencies for developers should be listed in requirements.txt.
• Alternatively, consider using http://pipenv.readthedocs.io.
• Package dependencies for users of your code (who will probably install via pip) should be listed in setup.py.

Code style

• Follow PEP8: https://www.python.org/dev/peps/pep-0008/
• Use pycodestyle to automatically check for PEP8.

Type checking

• Consider using type hints: https://docs.python.org/3/library/typing.html
• Consider using http://mypy-lang.org.
• Consider verifying type annotations at runtime: https://github.com/RussBaz/enforce

Share your package

• Choose a license and add a LICENSE file.
• Share your package on PyPI. For this you can follow https://github.com/bast/pypi-howto.

Documentation

• Use RST for your README (to make it easier for https://pypi.org).

Suggestions? Corrections? Pull requests?

Yes please!