Databricks Machine Learning Associate Question 480
Select 3An organization is developing a machine learning pipeline for a recommendation system. The development team frequently updates feature engineering code and tests different algorithms, while the production environment requires stability and consistency. Under which circumstances should the team prioritize promoting code over promoting models, and vice versa?
- A
Promote code when the feature engineering logic is frequently updated and needs to be tested in production.
- B
Promote code when the team wants to ensure the production environment is in sync with the latest changes in the pipeline's logic.
- C
Promote models when a finalized version has been thoroughly evaluated and is ready for deployment.
- D
Promote models when the production system needs to test multiple versions of the pipeline logic simultaneously.
- E
Promote code when the production environment requires stability and consistency over time.
Show answer and explanation
Correct answers: A, B, C
Explanation
The decision to promote code or models depends on the context. Promoting code is beneficial when the pipeline logic (e.g., feature engineering or preprocessing) is under active development and needs to be reflected in production. On the other hand, promoting models is better suited for scenarios where a finalized and validated model is ready for deployment, ensuring stability and consistency in production. Understanding these trade-offs is critical for managing machine learning workflows in Databricks.
- A. Correct.
Correct. Promoting code is appropriate when changes to feature engineering or pipeline logic are frequent and need to be reflected in production for testing or experimentation.
- B. Correct.
Correct. Promoting code ensures that the latest changes in the pipeline logic are mirrored in production, which is useful during active development and debugging.
- C. Correct.
Correct. Promoting models is ideal when a model has been finalized and validated, as this approach focuses on deploying a stable artifact without requiring immediate changes to pipeline logic.
- D. Incorrect.
Incorrect. Promoting models does not directly address testing multiple versions of the pipeline logic. Instead, this would involve techniques like A/B testing or using feature flags.
- E. Incorrect.
Incorrect. Promoting code is typically not aligned with ensuring stability in production, as it introduces changes which may disrupt the environment. Promoting models is a better choice for stability.