Feature Store Milestones

Nathalia Ribeiro Ariza
Feature Stores for ML
3 min readAug 31, 2021

This article is the result of a collaboration with Jim Dowling.

The feature store space is booming! From the introduction of the feature store concept by Uber in 2017, to the launch of the first Enterprise-ready feature store, by Logical Clocks the following year, to the entrance of tech giants Google, Amazon and Databricks, the need for platform support for feature management has never been clearer.

Featurestore.org was born from the realization that the list of in-house and Enterprise-ready feature stores was more extensive than one might have thought, but there was no single place on the Internet to learn about feature stores. Our mission is to give visibility to these great platforms and to make sure that you, our reader, are kept up to date on the latest and greatest in feature stores. Here, we summarize the most important feature store milestones, as we see it. This is far from an exhaustive list, so feel free to reach out to us and suggest improvements.

Designed with Freepik

What’s next for feature stores?

As feature store platforms mature, we anticipate they will gain new capabilities to improve the computation of features and management of data for AI. Here are some of our predictions on what will become increasingly important in the coming months and years:

  • Custom Metadata: an essential mechanism to describe and aid the discovery of features.
  • Provenance: platform support for tracking the dependencies between features, training data created from features, and the models that use features.
  • Tighter Serving Integration: automate and ease the burden on application developers who need to perform real-time transformations and join features from the online feature store, before sending feature vectors to online models for predictions.

Feature Store Summit

We want to bring visibility to great data platforms and discuss the latest and greatest in feature stores. Attend the Feature Store Summit, on October 12–13, to discover the latest technologies, best practices, use cases for putting ML models into production environments in the upcoming.

References

--

--