Machine Learning

Overview

New MLOps (DevOps for Machine Learning) capabilities in Azure Machine Learning brings the sophistication of DevOps to data science, with orchestration and management capabilities to enable effective ML Lifecycle management.

Key Features

- Using Azure Resource Manager an AzureML Service Connection type can be created to access your artifacts in an AzureML workspace. <br/>
- By registering a new version of a model into an AzureML service workspace, a trigger can be configured to kick off a release pipeline. <br/>
- Azure DevOps server task for running published ML pipelines
- Azure DevOps tasks for AzureML specific actions such as model profiling and model deployment.

How to set up your service connection

Select AzureMLWorkspace for the scope level, then fill in the following subsequent parameters.

How to enable model triggering in a release pipeline

Go to your release pipeline and add a new artifact. Click on AzureML Model artifact then select the appropriate AzureML service connection and select from the available models in your workspace.

Enable the deployment trigger on your model artifact as shown here. Every time a new version of that model is registered, a release pipeline will be triggered.

How to use the Azure DevOps AzureML tasks

Run published pipeline server task

Leverage Azure DevOps agentless tasks to run Azure Machine Learning pipelines. Go to your build pipeline and select agentless job.

Next, search and add ML published Pipeline as a task.

Fill in the parameters. AzureML Workspace will be the service connection name, Pipeline Id will be the published ML pipeline id under the Workspace you selected. Fill in experimentation name and expected pipeline parameters to run the pipeline.

Save your changes, and now you can invoke your published pipeline within your build or release pipeline.

Model Profile(Beta):

This task provides the optimal CPU and memory needed to run your model. Go to your release pipeline, select or add a stage.

Next, click into the stage, then search and add AzureML Model Profile as a task.

Provide the service connection name in AzureML Workspace, the files necessary for scoring your model, sample input data your model will expect and the desired file path which will store the profiling results.

Model Deploy(Beta):

Go to your release pipeline, and select or add a stage. Next, search and add AzureML Model Deploy as a task.

Fill in the parameters, AzureML Workspace will be the service connection name, inference config file with contain all the necessary dependencies required for scoring your model. Select the model deployment target and deployment configuration file. Templates for the inference config file and deployment config file can be found here: https://github.com/microsoft/MLOps/tree/master/model-deployment.

Resources:

MLOps project examples Get started with AzureML What is MLOps Get started with ML Pipelines Create reusable workflows with AzureML and Azure DevOps

Have any issues or feedback? Let us know here - https://feedback.azure.com/forums/257792-machine-learning