A Feature Store is an operational data management system for storing and serving machine learning (ML) features. Data scientists and ML engineers can use a feature store to transform, store and retrieve machine learning features, making them available to both ML training and serving workloads.
In this talk we will focus on state of the art feature stores that can be deployed and used in Cloud or or on-premise environments. Depending on the performance requirements, feature stores can be backed by different data storage technologies, such as relational databases or key-value stores. Within Hazelcast we have engineered a backend for a feature store leveraging Hazelcast’s compute and storage capabilities to achieve millisecond-level latencies for the needs of real-time prediction serving at scale.
Resources: