Databricks Machine Learning Professional Question 52
Single answerYou are managing a machine learning project on Databricks. After training an initial version of your model, you register it in the Databricks Model Registry with the version number 1. You later improve the model with better hyperparameters and register a new version (version 2). However, you are not ready to deploy version 2 yet because it still requires further validation. What is the best way to manage the model lifecycle in this scenario?
- A
Serve version 2 immediately in production to collect real-world feedback.
- B
Promote version 1 to 'Production' stage while keeping version 2 in 'Staging' stage.
- C
Archive version 1 and promote version 2 to 'Production' stage.
- D
Keep both version 1 and version 2 in 'Staging' stage until version 2 is fully validated.
Show answer and explanation
Correct answer: B
Explanation
The best practice in model lifecycle management on Databricks involves using the Model Registry to clearly separate stable production models from models undergoing validation. By promoting version 1 to the 'Production' stage and keeping version 2 in 'Staging', you ensure that the production environment remains stable while you validate the updates to version 2. This approach minimizes risk while enabling a structured path for model promotion.
- A. Incorrect.
Serving version 2 immediately in production without full validation is risky and not recommended for managing the model lifecycle effectively.
- B. Correct.
Promoting version 1 to 'Production' while keeping version 2 in 'Staging' allows you to maintain stability in production while validating the improved model in staging.
- C. Incorrect.
Archiving version 1 and promoting version 2 to 'Production' would prematurely remove the validated version from active use, which is not ideal without confirming the new model's performance.
- D. Incorrect.
Keeping both versions in 'Staging' would delay deployment unnecessarily and fail to take advantage of the stable and validated version 1.