Tecton, the enterprise feature store company and a primary contributor to Feast, announced Feast 0.10, the first feature store that can be deployed locally in minutes without dedicated infrastructure. The new release makes it possible for data scientists to reap the benefits of a functionally complete feature store with no infrastructure overhead or maintenance. Feast has seen strong adoption to date with more than 1,700 GitHub stars and contributions from Agoda, Cimpress, Farfetch, Google Cloud, Tecton and Zulily.
‘The simplified architecture will integrate with our existing infrastructure with minimal effort. The feast will help us centralise our features which promotes consistency, reusability and visibility of our prediction data’, said David Frantz, a senior software engineer at Porch, a home services platform company that connects homeowners with quality home improvement, repair and maintenance professionals.’
Feature stores have emerged as a critical component of the infrastructure stack for machine learning (ML). They solve the hardest part of operationalizing ML: building and serving ML data to production. They allow data scientists to build better ML features and deploy these features to production quickly and reliably.
Also Read: It’s an App, it’s Multiservice, No it’s a Super App
Until today, feature stores have not been lightweight or flexible enough to address the needs of individual data scientists and small data teams. Feast 0.10 is delivered as a Python software development kit (SDK) that can be deployed locally in minutes. Furthermore, Feast 0.10 is modular and integrates with existing data stacks, eliminating the burden and requirement of deploying and maintaining dedicated infrastructure.
‘Fast follower enterprises need to digitally transform their businesses and compete in a volatile post-COVID-19 environment. ML can help them, provided they implement accurate ML models that avoid infrastructure complexity. Data scientists must transform mountains of data, distil the right features, then use those features to train and deploy models. Feast 0.10 offers an open-source feature store to support this, and inevitable retraining and redeployment when the data drifts–on top of existing infrastructure’, said Kevin Petrie, Vice President of Research at Eckerson Group.
‘We originally open-sourced Feast to share our feature store technology and accelerate the deployment of all ML-powered applications. Feast 0.10 is a major milestone towards making feature stores easy to adopt for data teams that are just getting started in their operational ML journey’, said Willem Pienaar, creator and an official committer of Feast and architect at Tecton.
