workshop

MLOps Workshop

Introduction

The MLOps workshop is an instructor-led workshop that provides guidance on an MLOps implementation in Azure. This workshop leverages Azure Machine Learning and Github Actions to implement a robust set of workflows to support machine learning models in production.

The core business problem revolves around predicting taxi fares in New York. This is based on an Azure Open Dataset sourced from here. The need to predict numerical values is a regression problem that is a common need for many enterprises across data sets in their organizations. For the purpose of this workshop, the key stages of exploring the data, engineering predictive features (data engineering) and model building (training, hyperparameter tuning, algorithm selection, etc.) will be assumed to be done and already codified in this Jupyter notebook. The core focus of the workshop will then be how to productionalize this code, lay the DevOps foundation, and support the best model in production.

Audience

• Customer data scientists
• ML engineers
• ML platform architects and managers
• ... and any other roles that require hands-on experience to support ML models in Azure

Goals

• Understand key elements of modern MLOps and how it helps improve and accelerate ML practices.
• Design experiments, deployment environments and MLOps pipelines in Azure Machine Learning.
• Get hands-on experience in building continuous integration and continuous deployment pipelines with new Azure ML vNext and Github Actions.

Structure

• Pre-Workshop Checklist
• Part 0: MLOps overview and environment setup
• Part 1: Structure code for fast iterative development
• Part 2: Use cloud scale compute and monitor experiment with Azure ML
• Part 3: Use github for version control and automation
• Part 4: Continuous integration (CI)
• Part 5: Continuous deployment (CD)
• Part 6: Observability 

Repo Structure

Note: This is the repository file structure from the repository root.

• README.md > Core README for the repository
• .github
  • actions > YAML files for Github Actions relating to AML job creation, and endpoint deployment
  • workflows > YAML files for Github Actions relating to unit tests and CI/CD workflows
• src
  • workshop
    • README.md > README file highlighting the workshop goals, steps and key audience
    • conda-local.yml > Third-party python dependencies for managing the conda virtual environment
    • core
      • data_engineering > Python and YAML files to support feature engineering
      • evaluating > Python and YAML files to support model evaluation based on specific model metrics
      • pipelines > YAML files to support creation of ML pipelines
      • scoring > Python and YAML files to support model deployment and scoring
      • training > Python and YAML files to support model training
    • data > Base datasets in parquet, with a Python file to load the data into the default datastore
    • documents > Setup scripts, and markdown files to support a hands-on workshop
    • infra > Setup scripts to support initial creation of the Azure Machine Learning infrastructure and resources
    • notebooks > Jupyter notebook containing all the code related to data exploration, cleansing, feature engineering and model creation

Workshop Scenario

Note: Read before starting the workshop!

Your team has been working on a new machine learning (ML) problem (predicting taxi fares in New York). The team has been performing exploratory work on the data and has come to a state where the model is solidified. Now, it is time to put a structure into the work so that the team can iterate faster toward building a fully functional solution. So far, team members have been working mostly with Jupyter notebooks on their personal compute.

To re-engineer this into a functional MLOps process, the following steps will be taken:

1. The code will be modularized (refactored into separate python modules) and parameterized (configured so it can be re-run with different values). This will lay the foundation for good software practices and allow multiple data scientists/engineers to work collaboratively on the code. (Later, we will reinforce DevOps practices around continuous integration and continuous deployment with specific workflows to support model training and evaluation. MLOps builds off a strong foundation in DevOps and looks to additionally manage the model and data lifecycles to support the best model in production.)
2. After successfully restructuring the Jupyter notebook and running the modules locally, your team will leverage Microsoft Azure to run the ML experiment at scale. They will take advantage of experiment tracking and model management capabilities in Azure ML to keep track of experiments. The team will then deploy the model as a rest endpoint for real time inferencing.
3. They will then setup a centralized version control in Github to keep track of project code and manage different feature development tracks and releases. They will need to understand how to automate and orchestrate common tasks such as environment setup, training, and testing.
4. After setting up GitHub for MLOps, your team will start automating the model training and evaluation process with a Continuous Integration (CI) pipeline.
5. After a successful run of the CI pipeline, your team will complete the process with a Continuous Delivery (CD) pipeline that will handle the deployment of the model without introducing any downtime in production (hot swap).
6. Now, head to Workshop Environment Setup: Part 0

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

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.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.