CI/CD automates testing and deployment on every change.
Continuous integration runs your tests whenever code is pushed.
Continuous deployment ships passing changes to production automatically.
For ML this should also cover data validation and model evaluation.
A model that fails its evaluation gate must never reach production.
CI/CD automates the path from a code change to a running system.
Continuous integration runs the tests on every push.
Continuous deployment ships what passes, without a person driving the release.
The value is catching problems in minutes rather than after they reach users.
For machine learning the pipeline has to cover more than code.
Data validation belongs in it, since a schema change upstream breaks a model silently.
Model evaluation belongs in it, with a threshold that blocks the deploy.
A model that fails its evaluation gate must never reach production.
That gate is what separates an ML pipeline from a plain software one.
Test the serving path too, not just the training code.
A model that trains beautifully and fails to load is a common outcome.
Keep the pipeline fast, since a suite that takes an hour gets skipped.
Deploy behind a flag or to a fraction of traffic first.
Automate the rollback as carefully as the deploy, because you will need it under pressure.
Reproducibility is the foundation, so pin dependencies and record data versions.
One question on this card. Every option explains itself once you pick, right or wrong.
22 concepts in this module, then a graded test with every answer explained.