Datasets as Feature Tables
You'll install the Feature Store and Online Inference dbdemo and explore three notebooks that cover feature engineering from basics to production patterns in ~15 min.
Prereqs: 14. MLOps, Unity Catalog foundations
What you'll build
A working Feature Store demo with three notebooks, sample travel data, and feature tables registered in Unity Catalog. The demo covers the full lifecycle from feature creation through online serving.
Official source (Demo Center): Feature Store and Online Inference
Steps
1. Install the demo from the dbdemos library.
Run the following cells in a new Python notebook.
%pip install dbdemos
dbutils.library.restartPython()
import dbdemos
dbdemos.install('feature-store', catalog='main', schema='dbdemos_fs_travel')
Swap main and dbdemos_fs_travel for any Unity Catalog catalog and schema where your user has CREATE privileges.
2. Explore the installed notebooks.
The demo installs three notebooks that build on each other:
| Notebook | What it covers |
|---|---|
01_Feature_store_introduction | Ingest data, create and register Feature Tables in Unity Catalog, use FeatureLookup to join features, train a model with the Feature Engineering Client. |
02_Feature_store_advanced | Point-in-time lookups to prevent data leakage, Online Tables for real-time serving, Feature Specs, Feature Serving endpoints. |
03_Feature_store_pipeline | Build and manage feature tables declaratively using a Lakeflow Declarative Pipeline. |
Feature engineering is a data engineering task — not an ML task. Notebook 03 shows the recommended approach: define feature tables as a Lakeflow Declarative Pipeline so they refresh automatically and stay governed in Unity Catalog. Start there before writing any model training code.
Watch the walkthrough
Troubleshoot
ModuleNotFoundError: dbdemos
Run %pip install dbdemos, then dbutils.library.restartPython() before importing.
Permission denied writing catalog or schema
Pick a catalog and schema where your user can create tables, or ask a metastore admin to grant CREATE.
Next
- Do next: 14. MLOps overview
- Learn why: 14. MLOps
- Reference: Feature Store and Online Inference (Demo Center), Feature tables in Unity Catalog