How the H2O Feature Store Manages ML Features
- Predictive AI
- H2O Feature Store
- H2O Driverless AI
This lesson explains how H2O's built in feature store manages feature metadata, versioning, storage, and serving as Kubernetes native services. It covers how features get created through registration from raw data, how feature sets can be derived from other features into reusable pipelines, and how a searchable catalog lets data scientists find and reuse a feature someone else already built. It also covers materializing features to offline storage for training and online storage for low latency inference, and how Driverless AI can register engineered features, or even a trained model itself, back into the store for others to reuse.
Key takeaways
- How feature sets can be derived from other existing features and composed into reusable pipelines instead of built from scratch
- Why a searchable catalog of projects and features lets data scientists find and reuse a feature someone else already registered
- How the feature store materializes the same feature to offline storage for training and online storage for low latency inference
- How Driverless AI can register a newly engineered feature, or even a trained model, back into the feature store for other users
- Why Driverless AI's custom recipes and the generative AI platform's MCP tools let the feature store connect to external stores
Start owning your AI