Google Professional Machine Learning Engineer Question 448
Select 3Google Cloud PlatformYou are developing a machine learning pipeline on Google Cloud that requires tracking both model and dataset versions to ensure reproducibility. Which strategies should you implement to achieve this goal effectively?
- A
Use Google Cloud Storage to store datasets and include version-specific metadata in object names or metadata fields.
- B
Use Vertex AI Model Registry to manage and version your trained models.
- C
Rely exclusively on naming conventions in your local file system to track versions of datasets and models.
- D
Integrate Vertex ML Metadata to track lineage and versioning of datasets and models throughout the pipeline.
- E
Use BigQuery to store raw datasets and automatically configure it for version tracking with built-in versioning features.
Show answer and explanation
Correct answers: A, B, D
Explanation
Reproducibility in machine learning pipelines depends on effectively managing both model and dataset versions. Using Google Cloud Storage with metadata for datasets, Vertex AI Model Registry for managing models, and Vertex ML Metadata for tracking lineage ensures robust versioning and traceability throughout the pipeline. Avoid relying on manual or error-prone practices such as local file system naming conventions, and understand the limitations of tools like BigQuery for versioning.
- A. Correct.
Correct: Google Cloud Storage allows you to store datasets and include version-specific metadata, providing an easy way to manage dataset versions.
- B. Correct.
Correct: Vertex AI Model Registry is specifically designed to manage and version models, making it an essential tool for ensuring reproducibility.
- C. Incorrect.
Incorrect: Relying solely on naming conventions in a local file system is error-prone and does not scale well, especially in collaborative or production environments.
- D. Correct.
Correct: Vertex ML Metadata enables tracking of lineage and versioning, offering a comprehensive solution for maintaining reproducibility throughout the pipeline.
- E. Incorrect.
Incorrect: While BigQuery is excellent for storing and querying datasets, it does not natively provide dataset versioning. Additional mechanisms would be needed to achieve this.