Workshop for Machine Learning on Azure
Originally forked from: hyssh/mtc-open-workshop
Welcome to Microsoft Technology Center workshop for an introduction to Machine Learning on Azure. You will learn the basics of how to get started with the different ways of doing machine learning tasks on Azure Machine Learning Service. Following are some of the pre-requisites for you to have the best learning experience in this workshop.
Pre-requisites
-
Basic knowledge of Cloud Computing, Python, and Machine Learning Concepts
-
You should have an Azure subscription with appropriate permissions to be able to perform the tasks in this workshop.
You may also be able to use credits available with the Visual Studio Subscription if you have one.
You will typically need Contributor permission to be able to create an Azure Machine Learning resource.
You can try to create an Azure Machine Learning Workspace before the workshop for a head start and make sure you have the premissions to do so.
If you are unsure about resource quotas available to you, you may want to check out workspace level quotas and request an increase. NOTE: Quota increase requests make take a few minutes or hours depending on your subscription support service levels.
Agenda
Day 1
Session 1: (10am - 12pm)
- Welcome and Introductions
- Machine Learning process and Azure
- Demo: Environment setup
-
Exercise: Setup your Azure resources (Set up Azure ML Service)
For this workshop, we recommend using the public endpoint networking option. You may create private endpoint but due to the scope of this exercise, we shall not troubleshoot networking issues during the workshop.
Before we move on to the next part of the workshop, please make sure you have access to the workspace you just created by going to ml.azure.com and get to your newly created workspace through the web interface.
Following resources may be useful for setting up software environments and private python packages:
Session 2: (1pm - 3pm)
- Automated Machine Learning
- Hands-On: Create your first AutoML model and test it
- No code AutoML exercise (classification)
- No code AutoML exercise (time series forecasting)
-
For a more advanced exercise, you can also try AutoML with the Python SDK
The python SDK has more advanced functionality, not available in the web based UI.
- Turn off unused resources
- You can delete compute instances without losing your work on the notebooks or experiment telemetry.
- Make sure compute clusters have a minimum number of nodes set to ‘0’ (and Idle seconds before scale down set to a reasonable number) so they can resize to zero when idle.
Day 2
Session 1: (10am - 12pm)
-
No-code/low-code AzureML Designer
- Hands-on: Try AzureML studio yourself
- Train a no-code regression model
- Deploy a real-time inference model from your no-code ML experiment
- AzureML using Python SDK
For local development: Setting up environment for notebooks on your local machine. For the workshop, we can use compute instances in AzureML Service.
Clone AzureML samples into a location in your files as shown in the following figure:
Session 2: (1pm - 3pm)
-
Hands-on: AzureML Python getting started exercise
-
Start with a very simple example: Quickstart: Train and deploy a model in Azure Machine Learning in 10 minutes
This notebook can also be found in: Samples > tutorials > compute-instance-quickstarts > quickstart-azureml-in-10min
All links to sample notebooks are on github, they can also be found within the AzureML workspace > Notebooks > Samples that can be cloned into your own files.
-
-
Optional advanced exercise: Training on a remote linux VM
-
Turn off unused resources
Day 3
Session 1: (10am - 12pm)
- Large Scale Machine Learning on Azure
- Deep Learning leveraging GPU
Session 2: (1pm - 3pm)
- MLOps capabilities wth Azure ML
- AzureML Pipelines
- Turn off unsed resources
- Survey & Feedback