Currently, there is a Java FeatureFetcher library which can be used directly in ML applications or be hosted in a standalone service.
However, many ML applications are written in Python, and hence it is important to have a Python library to directly fetch features online.
Other alternatives considered:
Have a Java proxy service to access features from online KVStore, and expose it with a RPC or REST endpoint.
Why not consider it: Many companies do not have JVM (Java/Scala) in their tech stack
Use Py4J
Why not consider it: While it is an viable option, Py4J has to serialize Python objects to Java and vice versa. This adds complexity when dealing with data types and objects and can add overhead due to the conversion process.
Currently, there is a Java FeatureFetcher library which can be used directly in ML applications or be hosted in a standalone service. However, many ML applications are written in Python, and hence it is important to have a Python library to directly fetch features online.
Other alternatives considered: