AI reading noteChip Huyen
Designing Machine Learning Systems
A systems view of machine learning after the notebook: data sources, deployment, feedback, monitoring, and retraining. A model only helps when the surrounding system can deliver and sustain its predictions.
Designing Machine Learning Systems
Key ideas
- Choose evaluation metrics that reflect the product decision.
- Expect data distribution and user behavior to change.
- Design monitoring and update paths before a model goes live.
How I would apply it
For each model, document the incoming data contract and a drift alert.