We provide the recommended general guidelines:
Each DS project should have at least two resource groups with with one AML instance in each: DEV-RG and PROD-RG. We recommend however to also setup a TEST-RG.
The code in DEV must be able to be run in TEST/PROD without applying any change to it
Data scientists (Dev team) have no right to modify code in TEST/PROD. They should have only view access rights.
Create a dev branch from main from which one can create sub branches (feature engineering, hot fixes, etc).
Continuous Integration consists in running: linting, unit tests, data checks,…, and finally publishing an AML pipeline (either training or serving).
Any change applied to the main branch must be done through a Pull request. The approved PR may trigger a continuous integration process in TEST/PROD.