Databricks Machine Learning Associate Question 479
Select 2A team is developing a machine learning pipeline using Databricks. They are debating whether to promote their model code or the trained model itself between environments (e.g., from staging to production). In which of the following scenarios should they promote the trained model rather than promoting the code?
- A
The model training process is computationally expensive and time-consuming, making retraining in the production environment impractical.
- B
The production environment has stricter security constraints, allowing only pre-approved code deployments.
- C
The trained model includes pre-computed embeddings or feature transformations that are difficult to replicate in production.
- D
The development team wants to ensure that the latest version of the training code is used to retrain the model in production.
- E
The team needs to audit how the model was trained by inspecting the code used to generate it.
Show answer and explanation
Correct answers: A, C
Explanation
Promoting trained models is preferred when the production environment cannot easily replicate the training process due to computational costs or the complexity of pre-computed artifacts like embeddings. On the other hand, promoting code is better when retraining in production is feasible, as it allows for better control, auditing, and adherence to stricter security practices.
- A. Correct.
Promoting the trained model is preferred when retraining is computationally expensive or time-consuming. This avoids unnecessary duplication of training efforts and ensures the exact model is available in production.
- B. Incorrect.
Promoting code, not models, is preferred in environments with strict security constraints, as it allows the production environment to control the retraining process and verify the code.
- C. Correct.
Promoting the trained model is beneficial when it includes pre-computed embeddings or transformations that are hard to replicate. This ensures consistency and avoids errors in reproducing the same state.
- D. Incorrect.
This scenario favors promoting code, as retraining in production ensures the most up-to-date and verified code is used to generate the model.
- E. Incorrect.
Promoting code is necessary for auditing purposes, as it allows the team to inspect and verify the code that was used to train the model.