title | description | icon |
---|---|---|
Wyvern Feature Store |
This page explains how Wyvern's feature store works. |
store |
Wyvern Feature Store currently is built on top of feast and provides the following responsibilities:
- Defining a Feature Registry
- Storing historic (date-stamped) feature data for model training
- Materializing the latest feature data for real-time model inference
- Feature serving: Wyvern provides a default feature store solution which is built on top of feast
Wyvern's feature registry is currently built on top of feast. If you're already using feast for
Wyvern uses feast to materialize the latest feature data for real-time model inference
Previously, we talked about how Wyvern pipeline logs real time features generated during inference, which is piped back to your data warehouse.
Wyvern's get_historical_features is built on top of feast's get_historical_features. It aggregates the offline features in your data warehouse (which is supported by feast's get_historical_features) and the historical real-time features being consumed by wyvern pipeline.
This interface returns a curated training feature dataset to use, which are wyvern request based.
Get the latest feature data