Google Professional Machine Learning Engineer Question 123
Select 3Google Cloud PlatformYou are tasked with building a machine learning pipeline for a recommendation system. To ensure feature consistency across training and serving, you decide to use Vertex AI Feature Store. During implementation, which of the following actions should you take to properly create and consolidate features in the Feature Store?
- A
Organize features into feature groups called Feature Entities based on logical groupings such as users or products.
- B
Ensure each Feature Entity has a unique entity ID and specify the corresponding entity type.
- C
Combine all features into a single Feature Entity to simplify management and avoid redundancy.
- D
Use timestamps to track feature values over time for point-in-time correctness.
- E
Store features directly in your ML model code to reduce dependency on external services.
Show answer and explanation
Correct answers: A, B, D
Explanation
To effectively use Vertex AI Feature Store, it is important to organize features into logical groups called Feature Entities and assign them unique IDs and types. Timestamps are crucial for ensuring point-in-time correctness, which prevents data leakage and maintains feature consistency across training and serving. Avoid practices like combining all features into one entity or hardcoding features into model code, as these reduce manageability and scalability.
- A. Correct.
Correct. In Vertex AI Feature Store, features are grouped into Feature Entities, which are logical groups like users or products, allowing for better organization and retrieval.
- B. Correct.
Correct. Each Feature Entity must have a unique entity ID and specify the entity type, as this is necessary for proper management and association of features.
- C. Incorrect.
Incorrect. Combining all features into a single Feature Entity would reduce clarity and scalability, making it harder to manage and query features effectively.
- D. Correct.
Correct. Using timestamps ensures point-in-time correctness, which is critical for preventing data leakage during training and serving.
- E. Incorrect.
Incorrect. Storing features directly in your ML model code is an anti-pattern as it reduces reusability, scalability, and consistency between training and serving.