Google Cloud’s Feast, a latest open source feature store that released last week, to assist businesses to better manage, store, and discovers new features for their Machine Learning projects. This new open-source feature is a collaboration project between Google Cloud and GO-JEK, and Indonesian tech startup. Feast is an open, extensible, and a unified platform for feature storage.
Senior Data Scientist at GO-JEK, Peter Richens stated that Feast is an indispensable component in developing end-to-end Machine Learning systems at GO-JEK. He further added that his company is very thrilled to release it to the open-source community. Feast built to identify solutions for general challenges faced by Machine Learning development teams, include Machine Learning features not being reused, Diversification of the feature definitions, Hard to serve up-to-date features, and Inconsistency between training and serving. So, Feast wards off these challenges by offering teams with a unified platform that enables teams to easily reuse the other teams’ feature across the various projects.
It also controls the ingestion of data by allying it from both batch and streaming sources utilizing Apache Beam into the feature warehouse and feature serving stores. So, users can then question features in the warehouse practicing the same set of feature identifiers. Further, Feast also enables easy access to old feature data for its users, which in turn, can be utilized to create datasets for training models. and for teams, it enables to capture documentation, metadata, and metrics regarding features, allowing teams to interact coherently about these features.