Databricks Machine Learning Professional Question 72
Select 2A data science team has trained multiple models for a predictive maintenance use case and wants to use Databricks Model Registry to manage these models effectively. Which of the following actions can be performed using the Model Registry?
- A
Track the lineage of experiments and datasets used to train the models.
- B
Transition a model version to different stages such as 'Staging' or 'Production'.
- C
Serve a registered model directly as an endpoint without deploying to a production environment.
- D
Archive outdated model versions to prevent their usage in production workflows.
- E
Compare the performance metrics of different registered models side-by-side.
Show answer and explanation
Correct answers: B, D
Explanation
Databricks Model Registry is designed to manage the lifecycle of machine learning models. It allows users to transition model versions between stages, archive models, and maintain a centralized repository for model management. However, it does not directly handle experiment lineage, serve models as endpoints, or provide tools for side-by-side comparison of metrics.
- A. Incorrect.
The Model Registry itself does not track the lineage of experiments or datasets; this is handled by MLflow's experiment tracking functionality.
- B. Correct.
The Model Registry allows users to transition model versions between stages like 'Staging', 'Production', and 'Archived', making it easier to manage the lifecycle of a model.
- C. Incorrect.
While registered models can be deployed to production, the Model Registry does not serve models directly as endpoints. Deployment requires additional steps and infrastructure.
- D. Correct.
The Model Registry provides functionality to archive outdated model versions, ensuring they are not used in production workflows while still retaining them for historical reference.
- E. Incorrect.
The Model Registry does not natively provide a side-by-side comparison of performance metrics. Comparing metrics typically requires custom dashboards or external tools.