Releases · microsoft/MLOpsPython

Code refactoring

Concise and clean folders structure aligning with the "bring-my-own-code" concept
Simple and clean Python code
Update to the latest AzureML SDK

Linting and Unit testing

PR and CI pipelines perform linting (with flake8) and unit testing of Python code
PR and CI pipelines perform data integrity tests

Parameterization - via @tarockey

AzDo pipelines and Python code are based on a set of variables defined in a Variable Group in an AzDo Project. * For local development the variables are grouped in .env.example file

Environment Setup

Infrastructure as Code (IaC) pipelines to create (iac-create-environment.yml) and remove (iac-remove-environment.yml) all required resources for the repo (e.g. resource group, ML workspace, container registry, etc.)

Pipeline agent as a Docker container

AzDo build pipelines are configured to run tasks in a Docker container having installed Python and all required dependencies
The Docker Image is publicly available https://hub.docker.com/_/microsoft-mlops-python
There is an AzDo pipeline (docker-image-pieline.yml) building and pushing the docker image

AzDo Build Pipelines

Separate pipelines for PR and CI events
- PR pipeline (azdo-pr-build-train.yml) is triggered when a Pull Request to master is created. Performs linting, data and unit testing only.
- CI pipeline (azdo-ci-build-train.yml) is triggered when the code is merged to master. Performs linting, data testing, unit testing, building and publishing an ML pipeline.
PR and CI pipelines use a pipeline template (azdo-base-pipeline.yml) handling common tasks such as linting and testing

Model Deployment with AML Deploy Model task

The release pipeline deploying a model to ACI and AKS environments leverages AML Model Deploy Task

Rewritten documentation

Provide feedback