Azure AI Machine Learning CLI (az ml)

Installation

Install the CLI using pip or conda. The most common method is via pip:

pip install azure-cli
az extension add -n ml

Verify the installation:

az ml -h

Authentication

Log in to Azure and set your default subscription:

az login
az account set --subscription <subscription-id>

Authenticate the ML extension to your workspace:

az ml workspace set -w <workspace-name> -g <resource-group>

Compute Management

Create a new compute cluster:

az ml compute create -n mycluster -t amlcompute --min-instances 0 --max-instances 4

List compute resources:

az ml compute list

Datasets

Register a dataset from a storage account:

az ml data create -n mydata -p "{'path':'https://myaccount.blob.core.windows.net/data/train.csv'}"

Download a dataset locally:

az ml data download -n mydata -f ./data/

Model Management

Register a model artifact:

az ml model create -n mymodel -p ./outputs/model.pkl

List registered models:

az ml model list

Deployments

Deploy a model as an online endpoint:

az ml endpoint create -n myendpoint -f endpoint.yml
az ml deployment create -e myendpoint -n blue -f deployment.yml

Test the endpoint:

az ml endpoint invoke -n myendpoint -i '{"data": [1,2,3]}'

Examples

Run an experiment from a YAML file:

az ml job create -f job.yml

Submit a Python script as a job:

az ml job submit -c mycluster -f script.py --environment azureml:python-3.8

Command Reference

Below is a filtered list of the most frequently used commands. For the full reference, see the CLI Commands page.