Setting up a local feature store

In the Implementing features with Feature Store notebook, 8_feature_store_feast.ipynb, you use Feature Store in local mode, with a SQLite-based registry and online store and a file-based offline store, which requires no external infrastructure.

The Fraud Detection tutorial is not intended for a production environment. The 8_feature_store_feast.ipynb notebook uses local mode as a fully functional alternative.

For information on using Feature Store in a production environment, see Configuring Feature Store.

Prerequisites

You have imported the workshop files into your JupyterLab environment as described in Importing the workshop files into the JupyterLab environment.
You have run the 1_experiment_train.ipynb notebook and it produces the artifact/scaler.pkl.

Procedure

In your JupyterLab environment, open the 8_feature_store_feast.ipynb file and follow the instructions in the notebook.

The notebook guides you through the following tasks:
1. Install the feast Python package.
2. Prepare a 10,000-row data subset with entity IDs and timestamps (required by Feature Store).
3. Configure a local Feature Store instance.
4. Define feature views and a feature service.
5. Apply the definitions to the Feature Store registry.
6. Materialize features to the online store.
7. Retrieve historical features (training use case) and verify data integrity.
8. Retrieve online features (inference use case).
9. Integrate Feature Store features with the fraud detection model.