Feature stores have emerged as a pivotal component in the modern machine learning stack. They solve some of the toughest challenges in data for machine learning, namely feature management, storage, validation, serving, and reuse.
However, many feature store solutions require a coordinated effort from multiple teams, come with a large infrastructure footprint, and have high integration costs and operational overhead. This large investment places feature stores out of reach for the individual data scientist. What’s needed is a pluggable lightweight feature store that is simple to get started with, with minimal infrastructure requirements.
In this talk we will introduce a major upgrade to Feast, the leading open source feature store. The new Feast can be deployed simply through a pip install, while eliminating the need to deploy or manage dedicated infrastructure. It allows data scientists to develop production-ready ML applications from their local machines in minutes using the tools they're already familiar with.
We'll look at a new lightweight abstraction to feature data that creates a unified view of features that decouples models from environment-specific infrastructure. By publishing model-centric logical feature definitions, data scientists can now build ML applications that depend on any data source, using their tools of choice, and deploy to their existing production infrastructure.