Databricks Machine Learning Professional Question 90
Single answerYou are managing a machine learning project on Databricks and have multiple versions of a model in the Model Registry. The production team has requested that only the latest 'Production' model version remains active, while older versions should be archived to avoid clutter. Additionally, any 'Staging' model versions older than 6 months should be deleted. Which approach should you use to automate this process?
- A
Use a Databricks Job that runs a custom notebook with MLflow APIs to transition, archive, and delete model versions based on the specified conditions.
- B
Manually transition, archive, and delete model versions using the Databricks UI on a weekly basis.
- C
Set up a Delta Live Table pipeline to manage the transition, archive, and delete logic for model versions.
- D
Use the Databricks REST API to create an automated process that periodically transitions, archives, and deletes model versions as per the requirements.
Show answer and explanation
Correct answer: A
Explanation
To automate the transition, archive, and deletion of model versions in the Model Registry, a Databricks Job running a notebook with MLflow APIs is the most efficient and practical solution. The notebook can be programmed to check and enforce the specific rules, such as keeping only the latest 'Production' version active and deleting old 'Staging' versions. Manual actions and unrelated tools like Delta Live Tables are either inefficient or unsuitable for this use case.
- A. Correct.
This is the correct approach as it uses a Databricks Job with a custom notebook to automate model lifecycle management using MLflow APIs. The notebook can implement the logic to check for 'Production' and 'Staging' versions and perform the required actions.
- B. Incorrect.
This is not correct as manual intervention is prone to errors, time-consuming, and not scalable for managing model lifecycle in a production environment.
- C. Incorrect.
Delta Live Tables are designed for data pipeline automation, not for managing the lifecycle of models in the Model Registry. This option is irrelevant to the scenario.
- D. Incorrect.
Although using the Databricks REST API is a viable approach, it would require significant custom implementation effort. A Databricks Job with a notebook is a more practical and straightforward solution.