Train with MLflow Projects in Azure Machine Learning (preview)

In this article, learn how to submit training jobs with MLflow Projects that use Azure Machine Learning workspaces for tracking. You can submit jobs and only track them with Azure Machine Learning or migrate your runs to the cloud to run completely on Azure Machine Learning Compute.

MLflow Projects allow you to organize and describe your code so other data scientists (or automated tools) can run it. MLflow Projects with Azure Machine Learning enable you to track and manage your training runs in your workspace.

Learn more about the MLflow and Azure Machine Learning integration.

Prerequisites

• Install the MLflow SDK mlflow package and the Azure Machine Learning azureml-mlflow plugin for MLflow:

  pip install mlflow azureml-mlflow
  

  Tip

  You can use the mlflow-skinny package, which is a lightweight MLflow package without SQL storage, server, UI, or data science dependencies. We recommend this package for users who primarily need the MLflow tracking and logging capabilities but not the full suite of features, including deployments.

• Create an Azure Machine Learning workspace. To create a workspace, see Create resources you need to get started. Review the access permissions you need to perform MLflow operations in your workspace.

• To do remote tracking, or track experiments running outside Azure Machine Learning, configure MLflow to point to the tracking URI of your Azure Machine Learning workspace. For more information on how to connect MLflow to your workspace, see Configure MLflow for Azure Machine Learning.

• To use Azure Machine Learning as backend for MLflow projects, you need the package azureml-core.

  pip install azureml-core
  

Connect to your workspace

If you're working outside Azure Machine Learning, you need to configure MLflow to point to your Azure Machine Learning workspace's tracking URI. For more information, see Configure MLflow for Azure Machine Learning.

Track MLflow projects in Azure Machine Learning workspaces

This example shows how to submit MLflow projects and track them in Azure Machine Learning.

1. Add the azureml-mlflow package as a pip dependency to your environment configuration file to track metrics and key artifacts in your workspace.

   conda.yaml

   name: mlflow-example
   channels:
     - defaults
   dependencies:
     - numpy>=1.14.3
     - pandas>=1.0.0
     - scikit-learn
     - pip:
       - mlflow
       - azureml-mlflow
   

2. Submit the local run and set the parameter backend = "azureml". This parameter adds support for automatic tracking, model capture, log files, snapshots, and printed errors in your workspace. In this example, the MLflow project you want to run is in the current folder, uri=".".

   • MLflow CLI
   • Python

   mlflow run . --experiment-name  --backend azureml --env-manager=local -P alpha=0.3
   

   View your runs and metrics in the Azure Machine Learning studio.

Train MLflow projects in Azure Machine Learning jobs

This example shows how to submit MLflow projects as a job running on Azure Machine Learning compute.

1. Create the backend configuration object. In this example, set the backend configuration to COMPUTE. This parameter references the name of your remote compute cluster that you want to use for running your project. If you include COMPUTE, the project is automatically submitted as an Azure Machine Learning job to the indicated compute.

   • MLflow CLI
   • Python

   backend_config.json

   {
       "COMPUTE": "cpu-cluster"
   }
   
   

2. Add the azureml-mlflow package as a pip dependency to your environment configuration file to track metrics and key artifacts in your workspace.

   conda.yaml

   name: mlflow-example
   channels:
     - defaults
   dependencies:
     - numpy>=1.14.3
     - pandas>=1.0.0
     - scikit-learn
     - pip:
       - mlflow
       - azureml-mlflow
   

3. Submit the local run and set the parameter backend = "azureml". This parameter adds support for automatic tracking, model capture, log files, snapshots, and printed errors in your workspace. In this example, the MLflow project you want to run is in the current folder, uri=".".

   • MLflow CLI
   • Python

   mlflow run . --backend azureml --backend-config backend_config.json -P alpha=0.3
   

   Note

   Since Azure Machine Learning jobs always run in the context of environments, the parameter env_manager is ignored.

   View your runs and metrics in the Azure Machine Learning studio.

Clean up resources

If you don't plan to use the logged metrics and artifacts in your workspace, you can't delete them individually. Instead, delete the resource group that contains the storage account and workspace, so you don't incur any charges:

1. In the Azure portal, select Resource groups on the far left.

   

2. From the list, select the resource group you created.

3. Select Delete resource group.

4. Enter the resource group name. Then select Delete.

Example notebooks

The MLflow with Azure Machine Learning notebooks demonstrate and expand upon concepts presented in this article.

• Train an MLflow project on a local compute
• Train an MLflow project on remote compute.

Next steps

• Track Azure Databricks runs with MLflow.
• Query & compare experiments and runs with MLflow.
• Manage models registries in Azure Machine Learning with MLflow.
• Guidelines for deploying MLflow models.