MLflow Model Registry Flashcards

Which MLflow API method is used to transition a registered model version to the 'Production' stage?

Answer: client.transition_model_version_stage()

The MlflowClient method `transition_model_version_stage()` is used to move a model version between stages like Staging, Production, or Archived.

In Unity Catalog-backed MLflow, model names follow which naming convention?

Answer: catalog.schema.model_name

Unity Catalog uses a three-level namespace: catalog.schema.model_name, such as main.ml_models.my_classifier.

What is the purpose of model aliases in the MLflow Model Registry?

Answer: To assign a mutable named reference to a specific model version

Aliases provide a mutable, human-readable pointer (e.g., 'champion') to a specific version, making it easy to update which version serves a role without changing calling code.

Which function registers a model that was logged during an MLflow run?

Answer: mlflow.register_model()

`mlflow.register_model(model_uri, name)` takes the run artifact URI and registers it under a given name in the Model Registry.

When loading a model from the registry using a Unity Catalog alias, which URI format is correct?

Answer: models:/catalog.schema.model@alias

The correct URI format uses `models:/` followed by the three-level name and `@alias`, e.g., `models:/main.ml.my_model@champion`.

What does setting `registered_model_name` in `mlflow.log_model()` accomplish?

Answer: It logs and simultaneously registers the model in the registry in one step

Passing `registered_model_name` to `mlflow.log_model()` both logs the model artifact and registers it in the Model Registry as a single operation.

Which stage is NOT a valid built-in stage in the legacy (Workspace) MLflow Model Registry?

Answer: Deprecated

The legacy MLflow Model Registry has three built-in stages: Staging, Production, and Archived; 'Deprecated' is not a standard stage.