Google Professional Machine Learning Engineer Question 120
Select 3Google Cloud PlatformYou are working on a machine learning project that predicts customer churn for a subscription-based service. You decide to use Vertex AI Feature Store to manage your features. Your team has identified features from different data sources, such as user demographics from a CRM system, subscription activity logs, and customer support interactions. How should you organize and consolidate these features in Vertex AI Feature Store to ensure efficient use and reusability across multiple ML models?
- A
Group related features into feature groups based on their source system to simplify management.
- B
Create a single feature store for all features to avoid duplication and centralize access.
- C
Use unique feature IDs to ensure that feature names across different models do not conflict.
- D
Register each feature with metadata, such as data types and descriptions, for better discoverability and governance.
- E
Avoid including features that are not relevant to the current ML model to reduce storage costs.
Show answer and explanation
Correct answers: A, C, D
Explanation
In Vertex AI Feature Store, organizing features properly is critical for efficient management and reuse across multiple ML models. Grouping features into logical feature groups (entities) simplifies management and ensures that features are available for related ML tasks. Using unique feature IDs prevents naming conflicts, and registering features with metadata enables better governance and discoverability. These practices align with the principles of effective feature engineering and management in Vertex AI Feature Store.
- A. Correct.
Grouping related features into feature groups (also called entities) based on their source or logical grouping (e.g., users, sessions) is a best practice in Vertex AI Feature Store. This approach simplifies feature management and supports use cases where features are reused across models.
- B. Incorrect.
Creating a single feature store for all features is not recommended as it may lead to unmanageable complexity and inefficiencies, especially as feature sets grow.
- C. Correct.
Using unique feature IDs ensures that features from different models or sources do not conflict. This is essential for managing a centralized feature store effectively.
- D. Correct.
Registering features with metadata, such as data types, descriptions, and expected ranges, improves discoverability and governance, facilitating collaboration and reuse of features.
- E. Incorrect.
Excluding features that are not currently relevant to a specific ML model is not a best practice. Features should be curated and stored for potential reuse across multiple models, regardless of immediate relevance.