A Roadmap for Turning Theory and Data Projects into Python Packages
summary
Implements a handmade Lasso algorithm, compares predictions to scikit‑learn (small mean absolute difference), packs it into a Python package named lasso_reg, builds wheel and sdist, and uploads version 0.1.1 to PyPI using build and twine. Uses pytest and hypothesis for tests, demonstrates editable installs, and shows faster workflows with PDM build & publish and isolated tool installs via PEX. Ends by serving the trained model with an mlserver wrapper around FastAPI so HTTP inference works locally.
AI-written summary · about
chapters (9)
- 0:00Workshop introduction
- 11:48Lasso regression overview
- 30:00Building Lasso package
- 53:22Project metadata setup
- 1:05:00Versioning and PyPI
- 1:21:15Installing and testing
- 1:31:56Serving with ML Server
- 1:46:34Package with PDM
- 2:00:18Q&A
related talks
28mModern Python packaging without pain
01h 18mBuild Data Apps by Deploying ML Models as API Services
30mCreate a Python Package From Zero to Hero
01h 09mTurn Python Scripts into Beautiful ML Tools
01h 51mWrite your first package using literate programming
28mGoing From Notebooks to Scalable Systems
02h 36mPython Packaging Demystified
01h 28mMaintainable code in data science