This repo contains samples of machine learning (ML) projects we have often seen in the industry. For example, using image classification or object detection for quality control and safety monitoring. We are using these samples to showcase how to build, deploy, and operationalize ML projects in production with good engineering practices such as unit testing, CI/CD, model experimentation tracking, and observability in model training and inferencing.
Samples in this project leverage the basic ideas used in MLOpsPython. While MLOpsPython lays the foundation for operationalizing ML, we aim to provide representative samples and docs to
- provide a sounding startpoint to build a production quality ML solution of certain frameworks such as Tensorflow or Kaldi.
- demonstrate approaches and techniques for cross-cutting concerns such as unit testing and logging.
- document how to solve some of the challenges encountered in building ML solutions, such as security, data management, local vs. cloud based development and more.
This repo contains sample code and definition of Azure DevOps pipelines for CI/CD. These pipelines run in this Azure DevOps project. ML pipelines run in the Azure environment deployed using the Azure DevOps pipelines.
See the details on Repository Structure
Check out the samples and the docs. Once you are ready to run the code, follow these steps:
- Clone the repo.
- If you are only interested in one sample,
- if your dev machine is capable of running VSCode dev container, you can open the folder of the sample in VSCode, and reopen in a container as VSCode prompts you. Once VSCode builds the container, you can do all the development inside that container from that point on.
- if you don't want to develop in a container, you can go to the samples's
local_development
folder and run a script to create a conda environment for development. Refer to each sample's README for details.
- You can also open the entire project in VSCode. Refer to the README of the sample you are interested in, create a conda environment, and pick the Python interpreter from the environment in VSCode.
- Set up an Azure DevOps project to connect to your source repository.
Configure the pipelines using their yaml definitions in this repo:
- start with common/infrastructure to create an Azure environment to run the samples.
- follow README of the sample of interest to create the Azure DevOps pipelines. The pipelines are configured for continuous integration by default, so they automatically kick off for pull request validations and merging into the main branch.
- if a pipeline is triggered after a dependent pipeline, the dependent pipeline must be named unique in the entire Azure DevOps project, not just in a pipeline folder.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.