Quickstart: Install and get started with Azure Machine Learning service
Azure Machine Learning service (preview) is an integrated, end-to-end data science and advanced analytics solution. It helps professional data scientists prepare data, develop experiments, and deploy models at cloud scale.
This quickstart shows you how to:
- Create service accounts for Azure Machine Learning service
- Install and log in to Azure Machine Learning Workbench.
- Create a project in Workbench
- Run a script in that project
- Access the command-line interface (CLI)
As part of the Microsoft Azure portfolio, Azure Machine Learning service requires an Azure subscription. If you don't have an Azure subscription, create a free account before you begin.
Additionally, you must have adequate permissions to create assets such as Resource Groups, Virtual Machines, and so on.
- Windows 10 or Windows Server 2016
- macOS Sierra or High Sierra
Create Azure Machine Learning service accounts
Use the Azure portal to provision your Azure Machine Learning accounts:
Select the Create a resource button (+) in the upper-left corner of the portal.
Enter Machine Learning in the search bar. Select the search result named Machine Learning Experimentation.
In the Machine Learning Experimentation pane, scroll to the bottom and select Create to begin defining your experimentation account.
In the ML Experimentation pane, configure your Machine Learning Experimentation account.
Setting Suggested value for tutorial Description Experimentation account name Unique name Enter a unique name that identifies your account. You can use your own name, or a departmental or project name that best identifies the experiment. The name should be 2 to 32 characters. It should include only alphanumeric characters and the dash (-) character. Subscription Your subscription Choose the Azure subscription that you want to use for your experiment. If you have multiple subscriptions, choose the appropriate subscription in which the resource is billed. Resource group Your resource group Use an existing resource group in your subscription, or enter a name to create a new resource group for this experimentation account. Location The region closest to your users Choose the location closest to your users and the data resources. Number of seats 2 Enter the number of seats. Learn how seating impacts pricing.
For this Quickstart, you only need two seats. Seats can be added or removed as needed in the Azure portal.
Storage account Unique name Select Create new and provide a name to create an Azure storage account. The name should be 3 to 24 characters, and should include only alphanumeric characters. Alternatively, select Use existing and select your existing storage account from the drop-down list. The storage account is required and is used to hold project artifacts and run history data. Workspace for Experimentation account IrisGarden
(name used in tutorials)
Provide a name for a workspace for this account. The name should be 2 to 32 characters. It should include only alphanumeric characters and the dash (-) character. This workspace contains the tools you need to create, manage, and publish experiments. Assign owner for the workspace Your account Select your own account as the workspace owner. Create Model Management account check Create a Model Management account now so that this resource is available when you want to deploy and manage your models as real-time web services.
While optional, we recommend creating the Model Management account at the same time as the Experimentation account.
Account name Unique name Choose a unique name that identifies your Model Management account. You can use your own name, or a departmental or project name that best identifies the experiment. The name should be 2 to 32 characters. It should include only alphanumeric characters and the dash (-) character. Model Management pricing tier DEVTEST Select No pricing tier selected to specify the pricing tier for your new Model Management account. For cost savings, select the DEVTEST pricing tier if it's available on your subscription (limited availability). Otherwise, select the S1 pricing tier. Click Select to save the pricing tier selection. Pin to dashboard check Select the Pin to dashboard option to allow easy tracking of your Machine Learning Experimentation account on the front dashboard page of the Azure portal.
Select Create to begin the creation process of the Experimentation account along with the Model Management account.
It can take a few moments to create an account. You can check on the status of the deployment process by clicking the Notifications icon (bell) on the Azure portal toolbar.
Install and log in to Workbench
Azure Machine Learning Workbench is available for Windows or macOS. See the list of supported platforms.
The installation might take around 30 minutes to complete.
Download and launch the latest Workbench installer.
Download the installer fully on disk, and then run it from there. Do not run it directly from your browser's download widget.
A. Download AmlWorkbenchSetup.msi.
B. Double-click on the downloaded installer in File Explorer.
A. Download AmlWorkbench.dmg.
B. Double-click on the downloaded installer in Finder.
Follow the on-screen instructions in your installer to completion.
The installation might take around 30 minutes to complete.
Installation path to Azure Machine Learning Workbench Windows C:\Users\<user>\AppData\Local\AmlWorkbench macOS /Applications/Azure ML Workbench.app
The installer will download and set up all the necessary dependencies, such as Python, Miniconda, and other related libraries. This installation also includes the Azure cross-platform command-line tool, or Azure CLI.
Launch Workbench by selecting the Launch Workbench button on the last screen of the installer.
If you closed the installer:
- On Windows, launch it using the Machine Learning Workbench desktop shortcut.
- On macOS, select Azure ML Workbench in Launchpad.
On the first screen, select Sign in with Microsoft to authenticate with the Azure Machine Learning Workbench. Use the same credentials you used in the Azure portal to create the Experimentation and Model Management accounts.
Once you are signed in, Workbench uses the first Experimentation account it finds in your Azure subscriptions, and displays all workspaces and projects associated with that account.
You can switch to a different Experimentation account using the icon in the lower-left corner of the Workbench application window.
Create a project in Workbench
In Azure Machine Learning, a project is the logical container for all the work being done to solve a problem. It maps to a single folder on your local disk, and you can add any files or subfolders to it.
Here, we are creating a new Workbench project using a template that includes the Iris flower dataset. The tutorials that follow this quickstart depend on this data to build a model that predicts the type of iris based on some of its physical characteristics.
With Azure Machine Learning Workbench open, select the plus sign (+) in the PROJECTS pane and choose New Project.
Fill out of the form fields and select the Create button to create a new project in the Workbench.
Field Suggested value for tutorial Description Project name myIris Enter a unique name that identifies your account. You can use your own name, or a departmental or project name that best identifies the experiment. The name should be 2 to 32 characters. It should include only alphanumeric characters and the dash (-) character. Project directory c:\Temp\ Specify the directory in which the project is created. Project description leave blank Optional field useful for describing the projects. Visualstudio.com GIT Repository URL leave blank Optional field. A project can optionally be associated with a Git repository on Azure DevOps for source control and collaboration. Learn how to set that up.. Selected workspace IrisGarden (if it exists) Choose a workspace that you have created for your Experimentation account in the Azure portal.
If you followed the Quickstart, you should have a workspace by the name IrisGarden. If not, select the one you created when you created your Experimentation account or any other you want to use.
Project template Classifying Iris Templates contain scripts and data you can use to explore the product. This template contains the scripts and data you need for this quickstart and other tutorials in this documentation site.
A new project is created and the project dashboard opens with that project. At this point, you can explore the project home page, data sources, notebooks, and source code files.
You can configure Workbench to work with a Python IDE for a smooth data science development experience. Then, you can interact with your project in the IDE. Learn how.
Run a Python script
Now, you can run the iris_sklearn.py script on your local computer. This script is included by default with the Classifying Iris project template. The script builds a logistic regression model using the popular Python scikit-learn library.
In the command bar at the top of the Project Dashboard page, select local as the execution target and select iris_sklearn.py as the script to run. These values are preselected by default.
There are other files included in the sample that you can check out later, but for this quickstart we are only interested in iris_sklearn.py.
In the Arguments text box, enter 0.01. This number corresponds to the regularization rate, and is used in the script to configure the logistic regression model.
Select Run to start the execution of the script on your computer. The iris_sklearn.py job immediately appears in the Jobs panel on the right so you can monitor the script's execution.
Congratulations! You've successfully run a Python script in Azure Machine Learning Workbench.
Repeat steps 2 - 3 several times using different argument values ranging from 0.001 to 10 (for example, using powers of 10). Each run appears in the Jobs pane.
Inspect the run history by selecting the Runs view and then iris_sklearn.py in the Runs list.
This view shows every run that was executed on iris_sklearn.py. The run history dashboard also displays the top metrics, a set of default graphs, and a list of metrics for each run.
You can customize this view by sorting, filtering, and adjusting the configurations using the gear and filter icons.
Select a completed run in the Jobs pane to see a detailed view for that specific execution. Details include additional metrics, the files that it produced, and other potentially useful logs.
Start the CLI
The Azure Machine Learning command-line interface (CLI) is also installed. The CLI interface allows you to access and interact with your Azure Machine Learning service using the
az commands to perform all tasks required for an end-to-end data science workflow. Learn more.
You can launch the Azure Machine Learning CLI from the Workbench's toolbar using File → Open Command Prompt.
You can get help on commands in the Azure Machine Learning CLI using the --help argument.
az ml --help
Clean up resources
The resources you created can be used as prerequisites to other Azure Machine Learning service tutorials and how-to articles.
If you don't plan to use the resources you created here, delete them so you don't incur any charges.
In the Azure portal, select Resource groups on the far left.
From the list, select the resource group you created.
Select Delete resource group.
Enter the resource group name, and then select Delete.
If you see the error message "Cannot delete resource before nested resources are deleted," you must delete any nested resources first. For information on how to delete nested resources, see this troubleshooting section.
You have now created the necessary Azure Machine Learning accounts and installed the Azure Machine Learning Workbench application. You have also created a project, ran a script, and explored the run history of the script.
For a more in-depth experience of this workflow, including how to deploy your Iris model as a web service, follow the full-length Classifying Iris tutorial. The tutorial contains detailed steps for data preparation, experimentation, and model management.
While you have created your model management account, your environment is not set up for deploying web services yet. Learn how to set up your deployment environment.