Google Professional Machine Learning Engineer Question 313
Select 3Google Cloud PlatformYou are leading a team responsible for deploying machine learning models in production using Google Cloud. To ensure proper tracking of model versions and seamless updates, you decide to set up a model registry. Which practices should you follow to effectively organize the model registry?
- A
Use unique versioning for each model version stored in the registry.
- B
Store only the latest production model in the registry to reduce storage costs.
- C
Include metadata such as model hyperparameters, evaluation metrics, and training data version for each model entry.
- D
Use a centralized registry that integrates with CI/CD pipelines for automated deployment.
- E
Delete older versions of models from the registry once they are no longer in use.
Show answer and explanation
Correct answers: A, C, D
Explanation
Proper organization of a model registry involves practices that ensure traceability, reproducibility, and seamless integration into the ML lifecycle. By versioning each model uniquely, including relevant metadata, and using a centralized registry with CI/CD integration, teams can effectively manage their models. Avoid practices like limiting the registry to the latest model or deleting older versions, as they hinder auditability and operational flexibility.
- A. Correct.
Using unique versioning for each model version allows clear tracking of changes, rollbacks, and comparisons between different versions.
- B. Incorrect.
Storing only the latest production model is not a best practice, as it prevents you from rolling back or analyzing previous versions in case of issues.
- C. Correct.
Including metadata provides essential context for understanding model performance, reproducibility, and auditability.
- D. Correct.
A centralized registry that integrates with CI/CD pipelines streamlines deployment workflows and ensures consistency across environments.
- E. Incorrect.
Deleting older versions is not recommended, as historical models may be needed for debugging, audit purposes, or compliance requirements.