Use this for last‑mile review. Pair it with the Syllabus.
1) The lifecycle (exam framing)
Problem framing → data → training → evaluation → deployment → monitoring → iteration.
Correct answers usually include:
- a baseline
- an evaluation plan
- a deployment/rollback plan
- governance (versioning + access boundaries)
2) OCI Data Science core objects (high yield)
- Projects: organize work and access.
- Notebooks: interactive development.
- Jobs: repeatable batch runs (training, scoring).
- Models: versioned artifacts.
- Deployments: managed endpoints with controlled access.
3) Evaluation traps to avoid
- Leakage (features built from the target)
- Overfitting (great train metrics, poor generalization)
- Wrong metric for the cost of errors
- Non-representative validation split
4) Deployment and operations (concept-level)
- Version everything (data, code, model).
- Monitor quality + drift + latency + cost.
- Prefer staged rollouts; keep rollback paths.
- Don’t log sensitive features/payloads.