Google Professional Machine Learning Engineer Question 453
Select 3Google Cloud PlatformYou are managing a machine learning workflow on Google Cloud, and your organization requires strict compliance with auditing policies. You need to ensure that every version of your model and the data used to train it can be traced back to its origin, along with any transformations applied. Which combination of tools or practices should you use to achieve this?
- A
Use Vertex AI Model Registry to track model versions and metadata.
- B
Implement BigQuery for storing raw and transformed datasets with query logs enabled.
- C
Use Dataflow to preprocess data, and enable Data Catalog for metadata management.
- D
Activate Stackdriver Logging to store all API calls related to model training.
- E
Use Cloud Functions to automatically delete older versions of datasets and models to save costs.
Show answer and explanation
Correct answers: A, B, C
Explanation
To ensure compliance and maintain model and data lineage, you need tools that can track versions, transformations, and metadata. Vertex AI Model Registry provides comprehensive model version tracking, BigQuery ensures data lineage through query logs, and using Dataflow with Data Catalog supports lineage by managing preprocessing and metadata. Together, these tools provide end-to-end traceability for both models and data.
- A. Correct.
Vertex AI Model Registry tracks model versions, metadata, and lineage, making it a critical tool for compliance and auditing.
- B. Correct.
BigQuery can store both raw and transformed datasets, and query logs help trace data lineage to meet auditing requirements.
- C. Correct.
Dataflow can handle data preprocessing steps, and enabling Data Catalog provides detailed metadata management, ensuring data lineage can be traced.
- D. Incorrect.
While Stackdriver Logging is useful for monitoring, it does not directly enable model or data lineage tracking.
- E. Incorrect.
Automatically deleting older versions may conflict with auditing requirements, as lineage could be lost, violating compliance policies.