Professional Data EngineerOperationalizing machine learning modelsHard
A media company is building a recommendation system for video content. They need to serve personalized recommendations to users in real-time based on their viewing history and preferences. The features for the recommendation model, such as 'user_watch_count_genre_X' or 'item_popularity_last_24h', need to be consistently computed and served with very low latency for both training and online inference. Which Google Cloud service is best suited to manage and serve these features?
- ACloud SQL
- BCloud Spanner
- CVertex AI Feature Store
- DBigQuery
Show answer & explanationAnswer & explanation
Correct answer: C. Vertex AI Feature Store
Vertex AI Feature Store is specifically designed to manage, serve, and share ML features with low latency for both training and online inference, ensuring consistency and preventing training-serving skew.
Why the other options are wrong
- A. Cloud SQL is a relational database and not optimized for the high-throughput, low-latency feature serving required by ML models.
- B. Cloud Spanner is a globally distributed, strong consistent database, but it's a general-purpose database and doesn't offer the ML-specific optimizations and managed feature engineering capabilities of a feature store.
- D. BigQuery is a data warehouse optimized for analytics and batch processing, not real-time, low-latency feature serving for online inference.
Vertex AI Feature Store
A managed service for storing, serving, and sharing machine learning features for training and online inference, ensuring consistency and low latency.
- Centralized repository for ML features.
- Supports both batch and online serving.
- Helps prevent training-serving skew.
Memory trick: Store your features artfully for ML success.