Azure Machine Learning Service Workspace provides an area to Train, Validate, Package, Deploy and Monitor Machine Learning Services.
To work with the Azure Machine Learning Service Workspace you will need Azure Machine Learning Service SDK for Python and PowerShell Machine Learning CLI.
The Machine Learning Service Workspace Product Team are releasing new functionality to the service every sprint ie, 2 weeks.
Machine Learning CLI commands are used in CI/CD tasks, recently tasks have become available in the Azure DevOps Market Place. The tasks are in Beta and don’t have all the functionality needed yet.
Initial sources for learning;
Required Azure Resources
Machine Learning Service Workspace
Key Vault
Application Insights
Storage Account
Compute Power
Juptyer Notebook VMs can be used for Compute Power to get started, supports AKS, Machine Learning Compute, Azure Databricks, Data Lake Analytics, HDInsight when more is needed.
Authoring in the Machine Learning Service Workspace is still in Preview and currently not ready for full CI/CD, issues listed below.
The Azure resources are deployed by ARM Template. As part of the Machine Learning Service Workspace setup it automatically creates the keys it needs in the Key Vault.
By design the next time the ARM Template is deployed, eg CI/CD it will remove the Machine Learning Service Workspace keys that are created automatically as they haven’t been added to IaC, which breaks the Machine Learning Service Workspace and Jupyter Notebook VMs.
This will generate 500 errors in the Machine Learning Service Workspace Logs and prevent the Jupyter Notebook from functioning, or new ones being created.
Issue will be addressed in future updates to the Machine Learning Service Workspace.
Deploying Jupyter Notebook VMs by ARM Template is not supported and won’t work. There are no plans to support this, so you must point and click to create the Jupyter Notebook.