Databricks ML Associate Practice Question 1: UC Model Alias Promotion
Databricks Certified Machine Learning Associate · Test 1 · question 1 of 10
A registered model prod.fraud.scorer in Unity Catalog has version 3 carrying the alias champion. A nightly batch job loads the model with the URI models:/prod.fraud.scorer@champion. Version 5, currently aliased challenger, has beaten version 3 in offline evaluation.
How should the team promote version 5 so the nightly job uses it on its next run, without editing the job's code?
Show answer and explanation
Answer:
An alias is a mutable, named pointer to exactly one version of a registered model. Calling MlflowClient().set_registered_model_alias("prod.fraud.scorer", "champion", 5) moves the alias from version 3 to version 5. Any workload that loads models:/prod.fraud.scorer@champion resolves to the new version the next time it loads the model, so the job needs no changes. Promoting a challenger to champion this way is the pattern Databricks recommends for models in Unity Catalog.
Option A fails because stages (Staging, Production, Archived) are not supported for models in Unity Catalog. Databricks recommends aliases for deployment status and the catalog/schema namespace for environments.
Option C creates a separate registered model. The job still points at prod.fraud.scorer@champion, which still resolves to version 3.
Option D is destructive and gets the behaviour wrong. An alias does not fall back to the newest version when its version is deleted, so the job's alias URI would stop resolving.
Reference: Manage model lifecycle in Unity Catalog