Databricks Machine Learning Associate Question 481
Select 2You are working on a collaborative machine learning project in Databricks. The team is deciding between promoting code or promoting models in different deployment scenarios. Which of the following scenarios would favor promoting the code rather than the model?
- A
The model requires frequent retraining due to rapidly changing data patterns.
- B
The trained model is already optimized and ready for deployment in a production environment.
- C
The team wants to maintain reproducibility and allow for experimentation with different hyperparameters.
- D
The deployment environment has strict latency requirements, and the model needs to be served immediately.
- E
The project involves a pre-trained model that does not need retraining and works across different environments.
Show answer and explanation
Correct answers: A, C
Explanation
Promoting code is typically preferred when the focus is on ensuring reproducibility, frequent retraining, or experimenting with hyperparameters. In contrast, promoting models is more suitable for scenarios where the model is already trained, optimized, and ready for deployment without further modifications.
- A. Correct.
Promoting code is preferred when the model requires frequent retraining because it ensures that the code used for training remains consistent and reproducible across environments. Promoting the model in this case would not address the need for retraining.
- B. Incorrect.
Promoting the model is better suited when it is already optimized and ready for deployment, as it avoids the need to rebuild or retrain the model in production.
- C. Correct.
Promoting code is beneficial for reproducibility and experimentation. By promoting code, the hyperparameter tuning and training process can be replicated in different environments, allowing for adjustments and further experimentation.
- D. Incorrect.
Promoting the model is preferred in this scenario because it allows the model to be served immediately without additional training or tuning, which is critical for latency-sensitive environments.
- E. Incorrect.
Promoting the model is appropriate when using pre-trained models that are finalized and do not need retraining, as they can be directly deployed across various environments.