Clean Machine Learning Code: Practical Software Engineering...
summary
Claims that applying Clean Code and Clean Architecture to ML cuts technical debt and keeps change costs low backed by concrete examples: an insurance-premium pipeline switching from logistic regression to gradient boosted trees, notebooks used as manual tests that take an hour, opaque names like OVH_v5, and deployed ad systems processing 100 to 200 terabytes per day and 40 to 50 billion impressions per day. Recommends loose coupling, single responsibility principle, and avoiding pipeline jungles with patterns (dependency injection, interfaces) and familiar tools like scikit-learn TensorFlow Airflow pandas numpy.
AI-written summary · about
chapters (7)
- 0:00Clean ML code intro
- 3:11Why clean code matters
- 8:28Five ML pipeline principles
- 15:43SOLID for ML systems
- 24:23ML technical debt
- 34:30Legal risks of bad ML code
- 37:00Q&A
related talks
37mClean Architecture: How to Structure Your ML Projects to Reduce Technical Debt
37mBest Practices for Coding in ML/DS - Techniques to Construct your Project
27mKeys to Building Machine Learning Systems
39mBack to Basics: Cutting Through the AI-Generated Code Noise
32mMaking Models Work: How an optimized ML lifecycle can drive better ROI
39mRight Code, Right Place, Right Time
42mMasterclass: The right path to ML
41mBetter Code for Data Science