Hopsworks provides object store connectors to enable Parquet tables to be an external feature group in Hopsworks. Parquet tables can also be a source for feature pipelines, with features stored in Hopsworks’ Online/Offline stores. Parquet tables can also be sinks for predictions or logs produced by inference pipelines.
You can extend your object store with real-time feature serving capability with Hopsworks. You can also centrally manage features stored in Parquet tables in object storage on Hopsworks, enabling them to be joined with features from other data sources. You can easily use Parquet tables as data sources in Python or Spark feature pipelines.