Simplicity in an ML API means choosing an abstraction that hides implementation details without becoming leaky. A small set of methods and parameters makes the system easier to learn and use, while selected customization keeps debugging possible. section 13.1
Balance a simple interface with operational control:
- Hide most handlers and parameters behind a focused interface instead of exposing their interdependencies. section 13.1
- Retain explicit customization where it supports debugging and avoids the problems of underconfiguration. section 13.1
- Support simplicity across integration with layered APIs, separated ML and IO components, client libraries, feature toggles, and fallbacks. section 13.6
A configurable predict interface
A better predict function gives callers optional time and seed parameters, supplies a default time when needed, and passes the resulting values to the model.
