Deploying Automated ML for Azure Machine Learning Models

Azure Machine Learning is a cloud service that enables Machine learning professionals to efficiently manage the machine learning project lifecycle in a secure environment. Azure Machine Learning provides a robust ecosystem that includes not just a number of tools to assist developers in building long-lasting apps and services, but also a foundation for successfully monitoring them.

Creating a Machine learning model by selecting right data for featurization and algorithm for training and testing can be a time-consuming and repetitive process. Automated machine learning (AutoML) can help in these situations by providing pre-built task types for various use cases.

In this post ,we will create an Automated ML for Azure ML models using an open dataset and deploy it as a web service to predict on new data.


  1. An Azure Subscription

Create an Azure Machine Learning Workspace

  • Login to Azure Portal
  • Search for “Machine Learning” and Click on Create
  • Provide NameRegion and Pricing Tier and Click on Review + Create
  • Azure Machine Learning workspace is created.

Create and Run Experiment in Azure ML Workspace

  • Select Automated ML under the Author section.
  • Click on + New Automated ML run.
  • Click on + Create Dataset and select from your local files. For this demo we will use open datasets.
  • Select US Consumer Price Index Dataset and Click on Next.
  • Provide Name and Click on Create
  • Select the Dataset and Click on Next
  • Provide Experiment Name and Target Column to Predict and Select Compute Instance. If compute instance is not available we need to create one by Clicking on + New
  • Provide Name and Change default selections if needed. Click on Create.
  • Once the Compute instance is create, select the instance and Click on Next.
  • Select the Machine Learning Task type for experiment and Click on Next. Note: Different Machine learning tasks serve different purposes, before based on your project needs select appropriate task type. For our demo we are using Regression Task Type.
  • Select Validation Type : Auto and Click on Finish.

View Experiment Run Details

Experiment details can be viewed by clicking Models , Output + Logs etc. Individual run details can be viewed under Child-runs

You can also test your models with a test dataset to evaluate generated models. This feature is currently under preview, and may change.

Deploy Machine Learning Model

  • Select Run 1 on completion of experiment.
  • Select the model listed in the Best model summary section.
  • Select Deploy
  • Provide below details and click on Deploy
    • Name: Name of deployment
    • Description: Deployment description
    • Compute type: Select the type of endpoint to deploy
    • Enable authentication: token-based/key-based authentication.

Machine learning engineers can use customise featurization under view featurisation setting to to customise your experiment as per your needs like selecting columns, changing column type and imputing missings values with specific values in the data set.

I would encourage you to explore other machine learning task types to enhance your knowledge of Automated ML.

Categories: Azure, Azure Machine Learning

Tags: , , ,

Leave a Reply