Machine learning operations (MLOps)
Azure Machine Learning capabilities that automate and accelerate the machine learning lifecycle
MLOps helps you deliver innovation faster
MLOps, or DevOps for machine learning, enables data science and IT teams to collaborate and increase the pace of model development and deployment via monitoring, validation, and governance of machine learning models.
Training reproducibility with advanced tracking of datasets, code, experiments, and environments in a rich model registry.
Autoscaling, powerful managed compute, no-code deploy, and tools for easy model training and deployment.
Efficient workflows with scheduling and management capabilities to build and deploy with continuous integration/continuous deployment (CI/CD).
Advanced capabilities to meet governance and control objectives and promote model transparency and fairness.
Resource center
Take a walk through of the end-to-end MLOps process.
Access videos and accompanying notebooks, code samples and documentation.
MLOps feature dive: Manage your assets, artifacts and code
MLOps feature dive: Create event driven machine learning workflows - Microsoft Channel 9 Video
See MLOps in action
Build ML pipelines to design, deploy, and manage model workflows
Build ML pipelines to design, deploy, and manage model workflows
How to automate a machine learning pipeline
Sam is a data scientist working for an online fashion retailer. They’ve recently started to use machine learning to provide timely and accurate recommendations, improving both customer satisfaction and sales.
After a recent catalog update, Sam’s been asked to update their brand recommendation model. But there’s a problem – nobody’s quite sure where the current model came from, thus everything will need to be redone.
He turns to Azure ML to build a reproducible and traceable workflow.
Sam is a data scientist at an online retailer, tasked with updating their recommendation model after a recent catalog update. Sam is looking for the model pipeline, but no one knows where the model came from.
Sam turns to Azure ML to build a traceable model workflow.
Integrate Azure ML into your existing machine learning tools
Sam leverages the Azure ML SDK to bring his existing Jupyter notebook-based work to Azure ML. It allows him to take advantage of the scale and collaboration capabilities within Azure ML. His training code, regardless of the ML framework used, alongside his training datasets can be versioned and linked to a traceable experiment.
Capture the metrics from every experiment run
Every time Sam executes a run, Azure ML tracks its metrics and logs it for review. This practically snapshots the inputs, like datasets, code & parameters, as well as the outputs, like logs, metrics & models for assessment and comparisons.
Use the Azure ML workspace to review models and experiments
While Sam continues to work in his notebook environment, his team - both now and in the future - can use Azure ML studio to easily trace back from the final model in the model registry through to each asset and step used to create it.
Create a reusable pipeline to deliver efficiency and consistency
With Azure ML pipelines, Sam creates a reusable ML pipeline to deliver a model training workflow and trigger it automatically upon changes in source data. It also delivers efficiencies by reusing the outcome of steps with unchanged inputs.
Build reproducible machine learning pipelines
By leveraging Azure ML, Sam has built a reproducible and traceable workflow. Each model that the team produces is now tied to an experiment run, giving the team the ability to review and iterate. With a reproducible pipeline, Sam can let Azure ML automate retraining while he focuses on the next big thing.
Azure ML helps you build an enterprise-grade machine learning pipelines through reproducibility and traceability. This leads to more consistent model delivery with less variability and increased fault tolerance.
By leveraging Azure ML, Sam has built a reproducible and traceable workflow. Each model that the team produces is now tied to an experiment run, giving the team the ability to review and iterate. With a reproducible pipeline, Sam can let Azure ML automate retraining while he focuses on the next big thing.
Azure ML helps you build an enterprise-grade machine learning pipelines through reproducibility and traceability. This leads to more consistent model delivery with less variability and increased fault tolerance.
Deploy rapidly with confidence, using auto-scaling, managed, distributed inference clusters
Deploy rapidly with confidence, using auto-scaling, managed, distributed inference clusters
How to deploy a web service endpoint for a model
Sam, a data scientist working for an online fashion retailer, has recently added Azure ML to his workflow to build better models. His automated pipeline just registered a final model, complete with attached experimental results showing improved performance.
The web development team isn't quite as excited though. They’re not sure how to embed a machine learning model in their web app and have asked for a web service instead. Sam’s not a web service expert so he turns to Azure ML to help add a web service endpoint to his model.
Sam, a data scientist at an online retailer, used Azure ML to create and register a model. His web dev team has now asked him for a web endpoint to call his model for inferencing. Sam uses Azure ML to create this web service endpoint.
Fetch the latest model and profile it for production
Sam can use Azure ML to deploy and manage the web service endpoints. First, he fetches the latest model from the registry and profiles it, to see what kind of resources it might need in production.
Push the model to an Azure Container Instance
With the profiling data in hand, Sam uses Azure ML to package the model into a container and deploy it to an Azure Container Instance for real-time or batch inferencing. Azure ML provides automatic schema generation and monitoring of the model once deployed.
Deploy to scalable, diverse endpoints and implement A/B testing
Azure ML makes it easy to deploy your models wherever they are needed, from the cloud to the edge. It can leverage any compute options like CPU, GPU & FPGA. With just a couple of lines of code, Sam can push the same model to a more scalable Azure Kubernetes Service instance and even exercise advanced A/B testing capabilities for new model validation.
Enable the web dev team to use the deployed model
Once the model is deployed to an Azure Kubernetes Service instance, the web dev team can access the web service endpoint to call from their application for real-time or batch inferencing needs. The model endpoint is still monitored for data drift and other metrics that can be audited as needed.
Azure ML allows models to be published wherever they are needed
With Azure ML, Sam has taken his open source model and quickly turned it into a production-ready web service ready for his development team to use. As a data scientist, he’s been able to keep his focus on model development, rather than having to upskill on the many technologies needed for service hosting (including Kubernetes, Flask, and/or Swagger).
Azure ML helps data scientists seamlessly transition their models into mature and fully featured web services, enabling them to be rapidly adopted in applications.
With Azure ML, Sam has taken his open source model and quickly turned it into a production-ready web service ready for his development team to use. As a data scientist, he’s been able to keep his focus on model development, rather than having to upskill on the many technologies needed for service hosting (including Kubernetes, Flask, and/or Swagger).
Azure ML helps data scientists seamlessly transition their models into mature and fully featured web services, enabling them to be rapidly adopted in applications.
Integrate with Azure DevOps and GitHub Actions to automate ML workflows
Integrate with Azure DevOps and GitHub Actions to automate ML workflows
How to integrate with your DevOps workflow
Sam, a data scientist working for an online fashion retailer, has adopted Azure ML to build and deploy his models. He’s recently released his model to a production instance of Azure Kubernetes Service. The web team has integrated it into their web app with promising results.
As it's now a key part of the web app, Rebecca, part of the operations team, wants to ensure model build and deployment is part of their DevOps workflows. Sam’s not entirely sure of how to accomplish that. He introduces Rebecca to Azure ML to determine how it can satisfy their requirements.
Sam, a data scientist, recently used Azure ML to build and deploy models. He just released a model to their production AKS instance. Rebecca wants to integrate it with their DevOps workflows. Sam points her to Azure ML.
Create a new release pipeline with Azure Pipelines
Rebecca creates a new Azure Pipeline to automate the deployment process. The first step of the pipeline accomplishes tasks like building the model with all dependencies and checking for errors. To trigger this pipeline automatically, Rebecca connects it to the Azure ML model registry. This ensures it can start whenever a new model or a new version of the model is registered.
Publish the model for validation
Rebecca adds a task to publish the model to an Azure Container Instance for validation before it goes to production. Rebecca uses the Azure CLI to interact with Azure ML. It gives her access to the same features inside her tools, as Sam has in his notebook. This step mimics integration tests in their app DevOps process.
Move the model to production upon proper validation
Rebecca adds a release pipeline task to deploy the model to production with Azure Kubernetes Service. Before this task gets activated, Rebecca configures a gate requiring a lead/manager to assess and approve before continuing. This ensures the pipeline doesn’t cause major disruptions is something wasn’t caught in the scripted validations.
Setup data drift detection to trigger retraining
As Rebecca sets up the automated pipeline to deploy the models, Sam has enabled data drift monitoring with Azure ML. By monitoring service usage, he can see if the inference requests in production start to deviate from the training dataset, indicating a need to re-train.
Monitor service for stability and performance
With Azure App Insights integration, Rebecca can also monitor various aspects of service operations like number of requests, load on the AKS cluster, response latency etc. while Sam monitors model performance and determine if there are optimizations to be made there.
Monitor service for stability and performance
Azure ML has seamlessly integrated into Rebecca’s existing DevOps workflow, from release automation through production monitoring, with native integrations into existing tools like Azure DevOps and Azure App Insights. And while Rebecca keeps an eye on service operations, Azure ML provides Sam with tools to monitor and manage model performance and optimize as necessary.
Azure ML helps data science become a first-class citizen in your organization’s existing DevOps process, leading to more reliable and robust machine learning projects.
CI/CD with Azure DevOps and Github actions Detect data drift Github repo for this demo
Azure ML has seamlessly integrated into Rebecca’s existing DevOps workflow, from release automation through production monitoring, with native integrations into existing tools like Azure DevOps and Azure App Insights. And while Rebecca keeps an eye on service operations, Azure ML provides Sam with tools to monitor and manage model performance and optimize as necessary.
Azure ML helps data science become a first-class citizen in your organization’s existing DevOps process, leading to more reliable and robust machine learning projects.
CI/CD with Azure DevOps and Github actions Detect data drift Github repo for this demo
Create better governance and cost management across your ML projects
Create better governance and cost management across your ML projects
How to enable governance around an ML model
Phil, the IT Manager at an online fashion retailer, is enthusiastic about the work the data science and operations teams have been doing with Azure ML to deliver on the promise of machine learning in a robust and repeatable way.
Phil has a problem – he’s responsible for their IT infrastructure but isn’t quite sure what’s in production, who has access, or how much it’ll cost. He’s looking at Azure ML to help him deliver on these key governance responsibilities.
Phil, the IT Manager at an online retailer, is excited about Sam and Rebecca’s work. But Phil is responsible for IT infrastructure and doesn't know how to ensure governance around the ML project. He’s looking at Azure ML for help.
Review models in production
Azure ML enables Phil to audit the production endpoints. He can navigate back to underlying models and experiments, alongside the outputs of the model explanations and fairness analysis. This ensures Phil can trace every step in the model creation process and enable auditing the models for any bias.
Check the audit logs for detailed model history
Within Azure, he can access the Azure ML audit logs. This complements the traceability found in models and experiments with a detailed log showing every action tracked with date and time trace and the respective user. These logs help Phil trace actions initiated within or by the Azure ML service, for organization wide accountability.
Secure the workspace using standard Azure IAM
Phil can use Azure IAM to manage access to the Azure ML workspace. Azure ML provides a range of roles out-of-the-box. It also allows custom roles to be created. These roles can be associated with specific tasks that users can perform and control access to resources or the ability to submit jobs etc.
Ensure accountability with Azure resources
Along with the cost management features in Azure, Azure ML exhibits accountability via resource quotas that support configurable workspace level limits. This helps Phil manage spending and ensure equitable resource allocation across teams. It also encourages accountability by respective project teams to manage within the resources allocated.
Achieve governance and compliance with your ML projects models
Phil is thankful that Azure ML not only supports the data science and operations teams, but also delivers the enterprise-grade governance capabilities he needs to stay on top of his responsibilities. He can see exactly what’s in production, as well as ensure organizational compliance through auditing, role-based access control, and quota management.
Azure ML provides the organizational controls essential for making machine learning projects successful and secure.
Monitoring Azure ML Plan and manage costs for Azure ML Github repo for this demo
Phil is thankful that Azure ML not only supports the data science and operations teams, but also delivers the enterprise-grade governance capabilities he needs to stay on top of his responsibilities. He can see exactly what’s in production, as well as ensure organizational compliance through auditing, role-based access control, and quota management.
Azure ML provides the organizational controls essential for making machine learning projects successful and secure.
Monitoring Azure ML Plan and manage costs for Azure ML Github repo for this demo
See how customers are delivering value with MLOps
Sze-Wan Ng: Director of Analytics & Development, TransLink"With MLOps capabilities in Azure Machine Learning, we've improved bus departure predictions by 74 percent, and riders spend 50 percent less time waiting."
Vijaya Sekhar Chennupati, Applied data scientist, Johnson Controls"Using the MLOps capabilities in Azure Machine Learning, we were able to increase productivity and enhance operations, going to production in a timely fashion and creating a repeatable process."
