10.6 CI/CD for machine learning
A bad model cannot reach production silently.
CI and CD for machine learning add a gate that code-only pipelines lack: automated model evaluation as a release check, so a worse model cannot ship silently even when every code test passes. Canary and shadow deployments plus a rehearsed rollback complete the loop. It sits in the production module's core. The gap it closes is precise — code can be correct while the model regresses, and only an evaluation gate notices the difference.
Work through these
GitHub Actions: lint, typecheck, test
Automated checks that run on every change: style, types and tests. These are the same for machine learning as for any other software and are frequently skipped here.
Automated model evaluation as a release gate
An automated evaluation that must pass before a model can be released. This is the machine learning addition to an ordinary pipeline, and it is the one that prevents silent regressions.
Deployment strategies: blue-green, canary, shadow
Releasing to a parallel environment, to a small share of traffic, or alongside the current version without serving its results. Each gives a different amount of warning.
Rollback plans and kill switches
A tested way back to the previous version, and a switch that turns the feature off. Both need to have been exercised before the day they are needed.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.