Stream feature view
StreamFeatureView is a type of feature view in Feast that allows you to define features that are continuously updated from a streaming source. It is designed to handle real-time data ingestion and feature generation, making it suitable for use cases where features need to be updated frequently as new data arrives.
Supported Compute Engines
Key Capabilities
Real-time Feature Generation: Supports defining features that are continuously updated from a streaming source.
Transformations: Apply transformation logic (e.g.,
feature_transformationorudf) to raw data source.Aggregations: Define time-windowed aggregations (e.g.,
sum,avg) over event-timestamped data.⚡ Tiling with Intermediate Representations: Enable efficient pre-aggregation with correct merging semantics for holistic aggregations like
avgandstd. This provides faster queries while maintaining mathematical accuracy. Learn more about tilingFeature resolution & execution: Automatically resolves and executes dependent views during materialization or retrieval.
Last updated
Was this helpful?