Professional Data EngineerOperationalizing machine learning modelsHard
A data engineering team is building a real-time anomaly detection system. The ML model requires low-latency feature serving for online predictions. Specifically, it needs to access precomputed features (e.g., average transaction amount in the last 5 minutes, number of failed logins in the last hour) that are constantly updated. Which Google Cloud service is best suited for managing and serving these time-sensitive features efficiently?
- ACloud SQL
- BBigQuery
- CVertex AI Feature Store
- DCloud Storage
Show answer & explanationAnswer & explanation
Correct answer: C. Vertex AI Feature Store
Vertex AI Feature Store is specifically designed to serve precomputed features for online prediction with low latency and consistency, and for batch training. It manages feature definitions, ingests data, and serves features efficiently, which is critical for real-time ML systems.
Why the other options are wrong
- A. Cloud SQL is a relational database, generally not designed for the scale and low-latency requirements of serving features for real-time ML inference.
- B. BigQuery is a data warehouse optimized for analytical queries, not for low-latency, point-in-time feature lookups for online predictions.
- D. Cloud Storage is object storage, suitable for large datasets but not optimized for low-latency, real-time feature serving.
Vertex AI Feature Store
A managed service within Vertex AI that allows for the centralized storage, serving, and management of machine learning features for both online (low-latency) and offline (batch) use cases.
- Ensures feature consistency between training and serving.
- Provides low-latency online serving for predictions.
- Manages feature definitions and lineage.
Memory trick: Feature Store delivers your model's ingredients instantly, fresh and consistent.