Due to https://ml-ops.org/ the engineering flow of a business feature service is from data through model to code. The model (aka module) describes how a specific question is answered – the business application value proposition – based on data and executed via code. Models are the way of packaging of the logic applied on data to answer a specific question. The package contains
- all data pre-processing (cleansing, filtering, etc.)
- configurations (aka parametrization) e.g., the bias of the linear equitation and
- algorithms e.g., linear regression.
The model package is often abstracted from
- programming languages e.g., Python and
- frameworks e.g., PyTorch simplifying the usage of algorithm (tools, interfaces, library)
serialized via specific model-formats like e.g., PMMT, PFA or ONNX, delivered via different
- deployment mechanism e.g., Docker container on a Kubernetes cluster and
- serving methods e.g., HTTP web service.