- Source Primary Data Sets: It serves as the source for primary data sets, which are essential for Featureform operations.
- Transform Data Sets: The Offline Store executes transformations to build data sets registered through Featureform. This is where the real transformation work happens.
- Build Point-in-Time Correct Training Sets: It constructs training sets while ensuring point-in-time correctness, a crucial aspect for training machine learning models. Featureform creates and runs the proper transformation to build the training set, but the offline store is the engine to run this transformation.
- Build Batch Features: For batch features, it acts as the source of data to be materialized into the inference store, where the most recent values are stored. Featureform handles getting the features into the inference store.
- Maintain Historical Back-fill for Streaming Features: In cases involving streaming features, the Offline Store maintains a log of historical feature values, allowing us to create point-in-time correct training sets. Featureform will handle keeping the inference store nad offline store in sync.