Feature Engineering Overview

Wyvern’s feature engineering solution is built on top of the open source feature store solution feast and provides the following functionalities:

1. Feature registry
2. Online features: storing feature data in cache real-time model inference
3. Near real-time materialization: once your feature data shows up in your data warehouse, materialization happens right after and update the online feature store real-time model inference
4. Historical features: it provides get_historical_features interface, which not only covers offline features from the original features table but also covers the real-time features logged back to your data warehouse.

Wyvern Feature Store

Similar to the Wyvern application, Wyvern Feature Store is also built with FastAPI. To run the feature store server:

from wyvern import generate_wyvern_store_app

feature_store_app = generate_wyvern_store_app(path="path_to_feature_repo")

Now run:

uvicorn features.main:app

APIs

oneline feature API

The default API path is /feature/get-online-features. You can also configure the path with the WYVERN_ONLINE_FEATURES_PATH environment var.

historical feature API

The default API path is /feature/get-historical-features. You can also configure the path with the WYVERN_HISTORICAL_FEATURES_PATH environment var.

Let’s go to the next page to learn more.