Azure Workspaces

Defining Workspaces in Azure

• A workspace serves as a hub for building and deploying models
• We can create workspaces in the Azure Machine Learning service
• We can access workspaces using Python in an IDE
• A workspace will store the model's compute target and any experiment objects that are required for each model build
• Storing these experiment objects will allows us to track runs and retrieve logs, metrics, outputs, and scripts easily

• A workspace is defined by the following properties:

  • Workspace name: Our desired name for the workspace
  • Subscription: An Azure subscription to own this resouce (i.e. Visual Studio)
  • Resource Group: An Azure resource that has been allocated policies or permissions
  • Location: A location where the workspace will be created, so we typically want to select a location near where the workspace will be used

Defining Pipelines in Azure

• A pipeline is a tool used to create and manage workflows during our model deployment process

• This could include the following:

  • Data manipulation
  • Model training and testing
  • Deployment phases
• We can build pipelines in the Azure Machine Learning service

• Here are a few reasons why we would want to build a pipeline:

  • Scheduling tasks and executions

    • Which frees up data scientists' time

  • Allocating compute targets

    • Which makes it easier to scale up or down

  • Reusing pipeline scripts

    • Which makes it faster to setup the process of retraining and scoring models

  • Recording and managing input, output, and data

    • Which makes it easier for improving models later

References

• Azure Data Science Certification
• Defining MLOps with Python
• Describing Model Registrey and Deployment
• Describing Pipelines in Azure
• Describing Azure HDInsight
• Comparing Azure HDInsight and Databricks