· advanced
Putting a Model into Production
The distance between a notebook and something a business depends on. Serving, versioning, the skew between training and serving, watching for drift, and deciding when to retrain.
Written in house, from the work rather than from a syllabus. Every example is one we would use ourselves, and every quiz question explains why its answer is right.
What you’ll learn
- Choose between batch and real-time serving from the decision being served
- Version a model together with its data and code so a result can be reproduced
- Recognise training and serving skew, and design it out
- Monitor inputs, outputs and business impact rather than accuracy alone
- Distinguish data drift from concept drift, and respond to each
- Release with shadow or canary traffic, and keep a rollback that works
Before you start
- · Machine Learning with scikit-learn, or equivalent
- · Some familiarity with how services are deployed