✳flâneur — a map of the web's best reading

Powering Feature Stores with ClickHouse

clickhouse.com · 3,901 words · saved by 1 readers

In this article, we dive into the world of MLOps and explore Feature Stores: the different types, why you may need one, and the main components. Considering this, we present how ClickHouse can be used to power a feature store and, thus, the model lifecycle, providing performance and flexibility. This blog is mostly introductory on the topic and designed to act as a precursor to subsequent examples of training ML models using features residing in ClickHouse. As evidence of the potential here, we do present our recent integration with Featureform - an open-source “virtual” feature store, which we will use for future examples. Before addressing what a feature store is, it might be helpful to clarify what a feature is. Put simply, a feature is some property of an entity that has predictive power for a Machine Learning (ML) model. An entity, in this sense, is a collection of features as well as a class or label representing a real-world concept. The features should, if of sufficient quality

Introduction # In this article, we dive into the world of MLOps and explore Feature Stores: the different types, why you may need one, and the main components. Considering this, we present how ClickHouse can be used to power a feature store and, thus, the model lifecycle, providing performance and flexibility. This blog is mostly introductory on the topic and designed to act as a precursor to subsequent examples of training ML models using features residing in ClickHouse. As evidence of the potential here, we do present our recent integration with Featureform - an open-source “virtual” feature

Explore this link on the map →

related reading