Google Professional Machine Learning Engineer Question 445
Select 3Google Cloud PlatformYou are designing a machine learning pipeline on Google Cloud for a team that frequently updates both the training dataset and the model architecture. The team wants to ensure traceability of models and datasets to improve reproducibility and identify issues during deployment. Which steps should you take to implement proper model and dataset versioning?
- A
Use Vertex AI Model Registry to register and track different versions of your models.
- B
Store datasets in Cloud Storage and maintain versioning by enabling the 'Object Versioning' feature.
- C
Manually record dataset versions and model metadata in a local spreadsheet for traceability.
- D
Implement a naming convention for datasets and models stored in Google Cloud Storage to indicate version numbers.
- E
Use a combination of Vertex ML Metadata and Artifact Registry to track lineage, datasets, and models.
Show answer and explanation
Correct answers: A, B, E
Explanation
Proper model and dataset versioning is essential for reproducibility and traceability in machine learning pipelines. Google Cloud provides tools like Vertex AI Model Registry, Cloud Storage with Object Versioning, and Vertex ML Metadata to automate and streamline this process. These tools eliminate manual errors and ensure a scalable, robust approach to managing versions of datasets and models. While naming conventions or manual tracking might offer basic organization, they lack the comprehensive traceability and automation provided by Google Cloud's managed services.
- A. Correct.
This is correct. Vertex AI Model Registry is specifically designed to manage and trace different versions of machine learning models in a centralized repository, making it an essential tool for versioning and traceability.
- B. Correct.
This is correct. Enabling 'Object Versioning' in Cloud Storage allows you to automatically retain previous versions of your datasets, which is crucial for dataset versioning and reproducibility.
- C. Incorrect.
This is incorrect. Manually recording versions in a local spreadsheet is error-prone and not scalable for managing large-scale ML pipelines. Cloud-native tools should be used for automation and reliability.
- D. Incorrect.
This is incorrect. While naming conventions can help with organization, they are not sufficient for robust versioning or traceability, as they lack integrated tracking and metadata capabilities.
- E. Correct.
This is correct. Vertex ML Metadata tracks lineage and relationships between datasets, models, and other artifacts. Artifact Registry can further help manage datasets and model binaries, ensuring proper versioning and traceability.