Prerequisites

This tutorial assumes that you have an Azure Machine Learning service provisioned. If you do not have one yet, please follow this tutorial. This repo uses the following defaults:

• Workspace name: aml-demo
• Resource group name: aml-demo
• Workspace edition: Enterprise

Depending on your preference, you can either work directly on your local laptop using Docker (Path 1), or remotely connect to a Compute Instance (Path 2). Both paths will work, but depending on your system restrictions (e.g., not full admin rights to your laptop), you might want to use Path 2.

Both paths cover how to leverage Azure Machine Learning for performing demanding, long-running training and batch-scoring jobs in the cloud.

Path 1 - Using your Local Laptop

Follow this path if:

• You want to work fully locally on your laptop
• You can fullfil all these prerequisites:
  • You have full admin rights on your laptop
  • You have docker installed (see here)
• Note: This path does not support local testing using GPUs

Some parts might already be fulfilled (editor, SSH keys), hence feel free to skip those steps.

1. Execute training Python code on local laptop using python or startup script to make sure it is working properly

2. Install Azure CLI
   • Follow the documentation here to install the Azure CLI.
   • macOs (using brew): brew update && brew install azure-cli
   • Windows: Download installer
3. Login to Azure CLI

    az login
   

4. Install the AZ ML CLI extension

    az extension add -n azure-cli-ml
   

Path 2 - Using a Compute Instance

Follow this path if:

• You want (or are required) to work with a cloud-hosted machine in the background
• You do not have full admin rights on your laptop and have challenges installing, e.g., docker
• Want the most proven way possible

Some parts might already be fulfilled (editor, SSH keys), hence feel free to skip those steps.

1. Execute training Python code on local laptop using python or startup script to make sure it is working properly

2. Setup local editor with remote extension (Visual Studio Code or PyCharm)
   1. Option 1 - Install Visual Studio Code with Remote Development extension pack
      • Install Visual Studio Code
      • Install Remote Development
   2. Option 2 - Use existing PyCharm installation
      • Install PyCharm (TODO)
      • Setup Remote development (TODO)
3. Generate SSH Key
   • Open Terminal
   • Generate SSH key using ssh-keygen -t rsa -b 4096
   • Open ~/.ssh/id_rsa.pub with an editor of your choice and copy the public key
4. Provision Compute Instance in Azure Machine Learning
   • Open Azure Machine Learning Studio UI
   • Navigate to Compute --> Compute Instances
   • Select + New
   • Select a Compute name (doesn't matter)
   • Select a Virtual Machine size
   • Enable SSH Access
   • Paste your public key from ~/.ssh/id_rsa.pub in SSH public key
   • Hit Create
   • Wait until instance has been provisioned
   • Select SSH under Application URI and copy the Login using local account command
5. Connect remotely to the Compute Instance
   • Start Visual Studio Code
   • Select Remote Explorer
   • Click + Add new SSH Target
   • Paste copied command ssh azureuser@*.*.*.* -p *
   • Save update to C:\Users\username\.ssh\config (Windows) or ~/.ssh/config (macOS, Linux)
   • Right-click the new SSH Targets entry and select Connect to Host in Current Window
6. Optional - Mount default Fileshare from Azure Machine Learning to local laptop
   • Open the Azure Portal
   • Navigate to your Azure Machine Learning's Resource Group
   • Select the Azure Machine Learning's Storage Account (named same as the Machine Learning service with some number, e.g., amldemo12345678)
   • Select File shares
   • Search for prefix code and select the matching file share
   • Click Connect
   • Follow the instructions to mount the share under Windows, Linux or macOS

Now that you have your environment up and running, we can move to the next section and migrate your training code to Azure.