MLflow and Azure Machine Learning

APPLIES TO: Azure CLI ml extension v2 (current) Python SDK azure-ai-ml v2 (current)

• v1
• v2 (current version)

MLflow is an open-source framework that's designed to manage the complete machine learning lifecycle. Its ability to train and serve models on different platforms allows you to use a consistent set of tools regardless of where your experiments are running: locally on your computer, on a remote compute target, on a virtual machine, or on an Azure Machine Learning compute instance.

Azure Machine Learning workspaces are MLflow-compatible, which means you can use Azure Machine Learning workspaces in the same way that you'd use an MLflow server. Such compatibility has the following advantages:

• We don't host MLflow server instances under the hood. The workspace can talk the MLflow API language.
• You can use Azure Machine Learning workspaces as your tracking server for any MLflow code, whether it runs on Azure Machine Learning or not. You only need to configure MLflow to point to the workspace where the tracking should happen.
• You can run any training routine that uses MLflow in Azure Machine Learning without any change.

Tip

Unlike the Azure Machine Learning SDK v1, there's no logging functionality in the SDK v2 and we recommend using MLflow for logging. Such strategy allows your training routines to become cloud-agnostic and portable, removing any dependency in your code with Azure Machine Learning.

Tracking with MLflow

Azure Machine Learning uses MLflow Tracking for metric logging and artifact storage for your experiments. When connected to Azure Machine Learning, all tracking performed using MLflow is materialized in the workspace you are working on. To learn more about how to instrument your experiments for tracking experiments and training routines, see Log metrics, parameters, and files with MLflow. You can also use MLflow to Query & compare experiments and runs with MLflow.

Centralize tracking

You can connect MLflow to Azure Machine Learning workspaces even when you are running locally or in a different cloud. The workspace provides a centralized, secure, and scalable location to store training metrics and models.

Capabilities include:

• Track machine learning experiments and models running locally or in the cloud with MLflow in Azure Machine Learning.
• Track Azure Databricks machine learning experiments with MLflow in Azure Machine Learning.
• Track Azure Synapse Analytics machine learning experiments with MLflow in Azure Machine Learning.

Example notebooks

• Training and tracking an XGBoost classifier with MLflow: Demonstrates how to track experiments by using MLflow, log models, and combine multiple flavors into pipelines.
• Training and tracking an XGBoost classifier with MLflow using service principal authentication: Demonstrates how to track experiments by using MLflow from compute that's running outside Azure Machine Learning. It shows how to authenticate against Azure Machine Learning services by using a service principal.
• Hyper-parameter optimization using Hyperopt and nested runs in MLflow: Demonstrates how to use child runs in MLflow to do hyper-parameter optimization for models by using the popular library Hyperopt. It shows how to transfer metrics, parameters, and artifacts from child runs to parent runs.
• Logging models with MLflow: Demonstrates how to use the concept of models instead of artifacts with MLflow, including how to construct custom models.
• Manage runs and experiments with MLflow: Demonstrates how to query experiments, runs, metrics, parameters, and artifacts from Azure Machine Learning by using MLflow.

Important

• MLflow in R support is limited to tracking experiment's metrics, parameters and models on Azure Machine Learning jobs. Interactive training on RStudio, Posit (formerly RStudio Workbench) or Jupyter Notebooks with R kernels is not supported. Model management and registration is not supported using the MLflow R SDK. As an alternative, use Azure Machine Learning CLI or Azure Machine Learning studio for model registration and management. View the following R example about using the MLflow tracking client with Azure Machine Learning.
• MLflow in Java support is limited to tracking experiment's metrics and parameters on Azure Machine Learning jobs. Artifacts and models can't be tracked using the MLflow Java SDK. As an alternative, use the Outputs folder in jobs along with the method mlflow.save_model to save models (or artifacts) you want to capture. View the following Java example about using the MLflow tracking client with the Azure Machine Learning.

Model registries with MLflow

Azure Machine Learning supports MLflow for model management. This support represents a convenient way to support the entire model lifecycle for users who are familiar with the MLflow client.

To learn more about how to manage models by using the MLflow API in Azure Machine Learning, view Manage model registries in Azure Machine Learning with MLflow.

Example notebooks

• Manage model registries with MLflow: Demonstrates how to manage models in registries by using MLflow.

Model deployment with MLflow

You can deploy MLflow models to Azure Machine Learning and take advantage of the improved experience when you use this type of models. Azure Machine Learning supports deploying MLflow models to both real-time and batch endpoints without having to indicate and environment or a scoring script. Deployment is supported using either MLflow SDK, Azure Machine Learning CLI, Azure Machine Learning SDK for Python, or the Azure Machine Learning studio portal.

Learn more at Guidelines for deploying MLflow models.

Example notebooks

• Deploy MLflow to Online Endpoints: Demonstrates how to deploy models in MLflow format to online endpoints using MLflow SDK.
• Deploy MLflow to Online Endpoints with safe rollout: Demonstrates how to deploy models in MLflow format to online endpoints using MLflow SDK with progressive rollout of models and the deployment of multiple model's versions in the same endpoint.
• Deploy MLflow to web services (V1): Demonstrates how to deploy models in MLflow format to web services (ACI/AKS v1) using MLflow SDK.
• Deploying models trained in Azure Databricks to Azure Machine Learning with MLflow: Demonstrates how to train models in Azure Databricks and deploy them in Azure Machine Learning. It also includes how to handle cases where you also want to track the experiments with the MLflow instance in Azure Databricks.

Training MLflow projects (preview)

You can submit training jobs to Azure Machine Learning by using MLflow projects (preview). You can submit jobs locally with Azure Machine Learning tracking or migrate your jobs to the cloud via Azure Machine Learning compute.

Learn more at Train machine learning models with MLflow projects and Azure Machine Learning.

Important

Items marked (preview) in this article are currently in public preview. The preview version is provided without a service level agreement, and it's not recommended for production workloads. Certain features might not be supported or might have constrained capabilities. For more information, see Supplemental Terms of Use for Microsoft Azure Previews.

Example notebooks

• Track an MLflow project in Azure Machine Learning workspaces
• Train and run an MLflow project on Azure Machine Learning jobs.

MLflow SDK, Azure Machine Learning v2, and Azure Machine Learning studio capabilities

The following table shows which operations are supported by each of the tools available in the machine learning lifecycle.

Feature	MLflow SDK	Azure Machine Learning CLI/SDK	Azure Machine Learning studio
Track and log metrics, parameters, and models	✓
Retrieve metrics, parameters, and models	✓	1	✓
Submit training jobs	✓ 2	✓	✓
Submit training jobs with Azure Machine Learning data assets		✓	✓
Submit training jobs with machine learning pipelines		✓	✓
Manage experiments and runs	✓	✓	✓
Manage MLflow models	✓3	✓	✓
Manage non-MLflow models		✓	✓
Deploy MLflow models to Azure Machine Learning (Online & Batch)	✓4	✓	✓
Deploy non-MLflow models to Azure Machine Learning		✓	✓

Note

• ¹ Only artifacts and models can be downloaded.
• ² Using MLflow projects (preview).
• ³ Some operations may not be supported. View Manage model registries in Azure Machine Learning with MLflow for details.
• ⁴ Deployment of MLflow models to batch inference by using the MLflow SDK is not possible at the moment. As an alternative, see Deploy and run MLflow models in Spark jobs.

Next steps

• Concept: From artifacts to models in MLflow.
• How-to: Configure MLflow for Azure Machine Learning.
• How-to: Migrate logging from SDK v1 to MLflow
• How-to: Track ML experiments and models with MLflow.
• How-to: Log MLflow models.
• Guidelines for deploying MLflow models.