Google Professional Machine Learning Engineer Question 119
Select 4Google Cloud PlatformYou are tasked with building a feature store for a recommendation system using Vertex AI Feature Store. The system must serve online predictions and ensure that features are updated in near real-time. Which steps should you take to create and consolidate features in Vertex AI Feature Store for this use case?
- A
Define an entity type that represents the user or item for which recommendations will be generated.
- B
Write a Dataflow job to ingest and update the features into the Feature Store in real-time.
- C
Use BigQuery to store features and directly query them during prediction requests.
- D
Set up Feature Monitoring in Vertex AI Feature Store to detect data drift and anomalies in feature values.
- E
Ensure that features have a consistent schema and are versioned for reproducibility.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
To create and consolidate features in Vertex AI Feature Store for online predictions, you need to define entity types, set up real-time ingestion (e.g., using Dataflow), monitor feature quality using Vertex AI Feature Store's monitoring capabilities, and ensure schema consistency and versioning. BigQuery, while powerful for analytics, is not designed for the specific needs of a feature store, such as low-latency feature serving.
- A. Correct.
Defining an entity type is an essential first step because it determines the structure of your feature data and is required for storing and retrieving features.
- B. Correct.
A Dataflow job is commonly used for real-time feature ingestion and updates, which is critical for online predictions that rely on fresh data.
- C. Incorrect.
BigQuery is not typically used as a feature store; Vertex AI Feature Store is designed specifically for feature storage and retrieval.
- D. Correct.
Feature Monitoring in Vertex AI Feature Store is important to ensure the quality of your features over time and to detect potential issues like data drift.
- E. Correct.
Features need a consistent schema and versioning to maintain reproducibility and compatibility with your models, especially in production environments.