Part 3 - Microsoft Azure

Welcome to the third and final part of Phase 2 for the Data Science stream! This part will involve you doing (almost) everything you've done in the first two parts, except in the cloud!

Microsoft's cloud computing platform is called Azure and this part will specifically focus on Azure Machine Learning, a service within Azure that allows you to do everything from running Jupyter notebooks and training machine learning models in the cloud to logging evaluation metrics across multiple iterations of model development and deploying models from within the Azure portal (i.e. the website you'll use to access Azure from your web browser) or from your own code.

We hope you find that doing data science in the cloud makes several things easier and faster to accomplish while unlocking some useful features that can help you as a data scientist deliver more value in less time!

This part is smaller since we recognise that you may have spent most of your time on the first two parts, and also Azure Machine Learning consists of a deep set of tools that could take weeks to cover in themselves. The tasks below correspond to the tools visible in your workspace in Azure Machine Learning Studio, please complete them with the help of the resources provided for each task.

Getting Started

1. Complete Getting Started with Azure Machine Learning

2. Choose one dataset from your two chosen datasets to use for this part and complete the tasks below. Make sure you have completed the first two parts and are using your processed dataset and trained model for this part.

Tasks

1. Read the comments and code in the example notebook that we have provided for you to practice deploying a model onto Azure, and adapt the code within this notebook for your own work in the next task (please use the example notebook in a local environment for your own work and not inside Azure Notebooks in order to avoid errors in model saving or deployment).

2. Go to Endpoints and use the notebook above to create a real-time endpoint for one of your trained models from part 2 - resources:

   • Deploy real-time machine learning services with Azure Machine Learning

   • Deploy a model to a managed online endpoint

   • Deploy and score a machine learning model by using an online endpoint

3. Complete at least ONE of the following to the best of your ability and time available:

   • Go to Models and Notebooks to run your Jupyter notebook/s for model training and evaluation, and register your trained model in Azure - resources:

     • Train a machine learning model with Azure Machine Learning
     • Run a training script as a command job in Azure Machine Learning

   • Go to Jobs to run a script for tuning hyperparameters - resources:

     • Tune hyperparameters with Azure Machine Learning
     • Perform hyperparameter tuning with Azure Machine Learning
     • Track model training with MLflow in jobs

   • Go to Automated ML to automatically train and tune models to find the best one via experiments with different features, algorithms, and parameters - resources:

     • Find the best classification model with Automated Machine Learning
     • Automate machine learning model selection with Azure Machine Learning
     • automl-with-azureml

Submission

To complete this part, please include the following items in your repository:

• An entry script (score.py) for your model
• A Jupyter notebook (.ipynb) containing:
  • Code that calls your real-time endpoint (like in Step 7 of the example notebook)
  • Any other code, written explanations, results, screenshots of your Azure workspace (e.g. hyperparameter tuning results, precision and recall, ROC/AUC curves), and/or anything else to explain the steps you took and the outputs you generated

Marking Criteria

1. Task Completion

   • Written a working entry script
   • Created a working endpoint that takes input data as specified by the student and produces an accurate result (either a label (classification) or value (regression))
   • Completed at least one of the tasks in the list above (marking for this will be based on a comparative evaluation against other students)

2. Presentation

   • Your notebook/s and the code within them must be well-commented so that someone looking at your notebook/s for the first time can easily understand and follow along with what you did
     • Note that "well-commented" means that you have provided easily understandable explanations for what the results of your code mean or what you have inferred from them
     • We may penalise notebooks that only contain code and nothing else.

Tips

• This part relies heavily on the Microsoft Learn modules from Week 5 of Phase 1! In particular, we recommend that you go through the following resources to support your work:
  • Quickstart: Get started with Azure Machine Learning
  • Azure Machine Learning Hands-On Exercises from Phase 1 Microsoft Learn modules
    • https://microsoftlearning.github.io/mslearn-azure-ml
    • https://microsoftlearning.github.io/mslearn-dp100
  • Units 2 and 3 of Introduction to Azure Machine Learning
  • Unit 6 of Explore Azure Machine Learning workspace resources and assets