This repository demonstrates how IBM Decision Services can leverage ML predictive models hosted as micro services.
Material aims at tackling 3 challenges:
- how to host ML models in a simple and portable form factor,
- how to provide SDKs to easily consume ML driven predictions from remote applications,
- with the benefit of such SDK and ML micro service how to combine business rules and predictions in a decision service project.
The technical proposal fits with a concept of operations based on 3 main roles and 4 steps:
- Step 1: A Data scientist elaborates an ML model in a data science tool.
- Step 2: A Data scientist exports an ML model serialized in pickle or joblib.
- Step 3: A developer takes the serialized ML model and hosts it as a microservice
- Step 4: A Business user creates a decision service in IBM Digital Business Automation that invokes the hosted ML model
The approaches combines Python for ML, Docker and OpenAPI.
This repository is composed of 3 main parts:
- an ML microservice: a micro service architecture to host ML models as REST APIs in a Docker container.
- an ML microservice sdk: a sdk to remotely get a prediction from the microservice and manage the ML models.
- an integration pattern for IBM Decision Services mixing rules and machine learning: an integration pattern and project samples to automate your decisioning by blending prescriptive and predictive logic.
- miniloan project that leverages business rules plus the micro ml sdk to automate the processing of loan applications.