noindex

Machine Learning Operations – MLOps | Microsoft Azure

Machine Learning Operations – MLOps | Microsoft Azure
























Azure Machine Learning capabilities that automate and accelerate the machine learning lifecycle













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.









Deliver innovation rapidly

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.









Build reproducible workflows and models


Reduce variations in model iterations and provide fault tolerance for enterprise-grade scenarios through reproducible training and models. Use datasets and rich model registries to track assets. Enable enhanced traceability with tracking for code, data, and metrics in run history. Build machine learning pipelines to design, deploy, and manage reproducible model workflows for consistent model delivery.




Easily deploy highly accurate models anywhere


Deploy rapidly with confidence. Use autoscaling, managed CPU, and GPU clusters with distributed training in the cloud. Package models quickly and ensure high quality at every step using model profiling and validation tools. Use controlled rollout to promote models into production.




Efficiently manage the entire machine learning lifecycle


Use built-in integration with Azure DevOps and GitHub Actions for seamlessly scheduling, managing, and automating workflows. Optimise model training and deployment pipelines, build for CI/CD to facilitate retraining, and easily fit machine learning into your existing release processes. Use advanced data-drift analysis to improve model performance over time.




Achieve governance across assets


Track model version history and lineage for auditability. Set compute quotas on resources and apply policies to ensure adherence to security, privacy, and compliance standards. Build audit trails to meet regulatory requirements as you tag machine learning assets, and automatically track experiments for CI/CD. Use the advanced capabilities to meet governance and control objectives and to promote model transparency and fairness.




Benefit from interoperability with MLflow


Build flexible and more secure end-to-end machine learning workflows using MLflow and Azure Machine Learning. Seamlessly scale your existing workloads from local execution to the intelligent cloud and edge. Store your MLflow experiments, run metrics, parameters, and model artefacts in the centralised Azure Machine Learning workspace.









Resource center





Manage your assets, artifacts, and code



Create event-driven machine learning workflows



CI/CD with GitHub Actions











Increase time to value with MLOps best practises

More organisations are using machine learning to build predictive insights and drive business outcomes. Learn how to use MLOps to accelerate the development and deployment of machine learning models and increase time to value.







Accelerate the ML lifecycle with Azure Machine Learning

Faster access to AI and machine learning insights is critical to enterprise success. Learn how to build and deploy machine learning solutions quickly and efficiently and streamline workflows with Azure Machine Learning.



















Build ML pipelines to design, deploy and manage model workflows




Build ML pipelines to design, deploy and manage model workflows



Build ML pipelines to design, deploy and manage model workflows


Automating a machine learning pipeline

Automating a machine learning pipeline

Sam is a data scientist working for an online fashion retailer. The company recently started using machine learning to provide timely and accurate recommendations, improving both customer satisfaction and sales.

After a catalogue update, Sam was asked to update the brand recommendation model—but nobody knows where the current model came from. He turns to Azure Machine Learning 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.










Select these points to progress demo

Select these points to progress demo

Use Azure Machine Learning with your existing tools

Sam uses the Azure Machine Learning SDK to move his Jupyter Notebook–based work to Azure Machine Learning, taking advantage of its scale and collaboration capabilities. Using his choice of framework, he versions his training code and datasets and links them to a traceable experiment.

Capture the metrics from every experiment run

Every time Sam executes a run, Azure Machine Learning tracks its metrics and logs it for review. This captures the inputs (such as datasets, code, and parameters) and outputs (such as logs, metrics, and models) for assessment and comparisons.

Review models and experiments in the workspace

While Sam continues to work in his notebook environment, his team uses the Azure Machine Learning studio to easily trace back from the final model in the model registry through each asset and step used to create it.

Create reusable pipelines for efficiency and consistency

Sam creates a reusable machine learning pipeline to deliver a model training workflow and trigger it automatically upon changes in source data. If needed, he can reuse the outcome of steps without changing the inputs.






Build reproducible machine learning pipelines

Build reproducible machine learning pipelines

Sam has built a reproducible and traceable workflow using Azure Machine Learning. Each model that the team produces is now tied to an experiment run for review and iteration. With a reproducible pipeline, Sam lets Azure Machine Learning automate retraining while he focuses on the next big thing.

Azure Machine Learning helps you build enterprise-grade machine learning pipelines through reproducibility and traceability. This leads to more consistent model delivery with less variability and increased fault tolerance.


Learn more about machine learning pipelines
Browse the GitHub repo for this demo

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.


Learn more about machine learning pipelines
Browse the GitHub repo for this demo








Deploy rapidly with confidence, using auto-scaling, managed, distributed inference clusters




Deploy rapidly with confidence, using auto-scaling, managed, distributed inference clusters



Deploy rapidly with confidence, using auto-scaling, managed, distributed inference clusters


Deploying a web service endpoint for a model

Deploying a web service endpoint for a model

Sam, a data scientist working for an online fashion retailer, has added Azure Machine Learning to his workflow to build better models. His automated pipeline just registered a final model with attached experimental results showing improved performance.

The web development team isn’t sure how to embed a machine learning model in the web app and has asked for a web service instead. Sam turns to Azure Machine Learning for help adding 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.










Select these points to progress demo

Select these points to progress demo

Fetch the latest model and profile it for production

Sam 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 Azure Container Instances

With the profiling data in hand, Sam uses Azure Machine Learning to package the model into a container and deploy it to Azure Container Instances for real-time or batch inferencing. After deployment, he takes advantage of the automatic schema generation and monitoring features in Azure Machine Learning.

Deploy to scalable, diverse endpoints and implement A/B testing

With just a couple lines of code, Sam pushes the same model to a more scalable Azure Kubernetes Service (AKS) instance and exercises advanced A/B testing capabilities for new model validation. He can now deploy models wherever they’re needed, from the cloud to the edge, using compute options including CPU, GPU, and FPGA.

Enable the web dev team to use the deployed model

After the model is deployed to an AKS instance, the web dev team accesses the web service endpoint to call from their application for real-time or batch inferencing needs. Sam monitors the model endpoint for data drift and other metrics, which can be audited as needed.






Publish models wherever they're needed

Publish models wherever they’re needed

Sam turned his open-source model into a production-ready web service for his development team. He kept his focus on model development rather than learning the many technologies needed for service hosting, including Kubernetes, Flask, and Swagger.

Azure Machine Learning helps you seamlessly transition your models into mature and fully featured web services, enabling them to be rapidly adopted in applications.


Learn how to deploy models with Azure Machine Learning

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 has 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.


Learn how to deploy models with Azure Machine Learning








Interoperate with Azure DevOps and GitHub Actions to automate machine learning workflows




Interoperate with Azure DevOps and GitHub Actions to automate machine learning workflows



Interoperate with Azure DevOps and GitHub Actions to automate machine learning workflows


Integrating with your DevOps workflow

Integrating with your DevOps workflow

Sam, a data scientist working for an online fashion retailer, used Azure Machine Learning to build and deploy his models. He recently released his model to a production instance of Azure Kubernetes Service (AKS), and the web team has integrated it into the web app with promising results.

Rebecca, who’s on the operations team, wants to ensure that model build and deployment is part of the DevOps workflows. Sam isn’t sure how to incorporate it, so he trains Rebecca on Azure Machine Learning and works with her to determine how it can satisfy both 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.










Select these points to progress demo

Select these points to progress demo

Create a new release pipeline with Azure Pipelines

Rebecca creates a pipeline in Azure Pipelines 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 automatically trigger this pipeline, Rebecca connects it to the Azure Machine Learning model registry, ensuring that it starts 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 Azure Container Instances for validation before it goes to production. She uses the Azure CLI to interact with Azure Machine Learning. It gives her access to the same features inside her tools that Sam has in his notebook. This step mimics integration tests in their app’s DevOps process.

Move the model to production upon validation

Rebecca adds a release pipeline task to deploy the model to production using AKS. Before this task is activated, Rebecca configures a gate requiring a lead or manager to assess and approve before continuing. This ensures that the pipeline doesn’t cause major disruptions if an issue 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

Using Application Insights, a feature of Azure Monitor, Rebecca monitors service operations such as number of requests, load on the AKS cluster, and response latency, while Sam monitors model performance to determine whether to make optimisations.






Create a streamlined model workflow

Create a streamlined model workflow

Rebecca incorporated Azure Machine Learning into her DevOps workflow, from release automation through production monitoring. While she keeps an eye on service operations, Sam monitors and manages model performance and optimises as necessary.

Azure Machine Learning helps elevate data science in your organisation’s DevOps process, leading to more reliable and robust machine learning projects.


Detect data drift (preview) on datasets

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 organisation’s existing DevOps process, leading to more reliable and robust machine learning projects.


Detect data drift (preview) on datasets








Improve governance and cost management across your machine learning projects




Improve governance and cost management across your machine learning projects



Improve governance and cost management across your machine learning projects


Enabling machine learning model governance

Enabling machine learning model governance

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 Machine Learning.

Although Phil is responsible for the company’s IT infrastructure, he isn’t sure what’s in production, who has access, or how much it’ll cost. He decides to use Azure Machine Learning to help him deliver on his 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.










Select these points to progress demo

Select these points to progress demo

Review models in production

Phil uses Azure Machine Learning to audit the production endpoints. He can navigate back to underlying models and experiments along with the outputs of the model explanations and fairness analysis. This ensures that he traces every step in the model creation process and enables model auditing for any bias.

Check the audit logs for detailed model history

Phil accesses the Azure Machine Learning audit logs. They complement the traceability found in the models and experiments with a detailed log, which tracks every action (date, time, and user). These logs help Phil trace actions initiated within or by the service for organisation-wide accountability.

Secure the workspace using Azure identity and access management

Phil uses Azure identity and access management solutions to manage access to the Azure Machine Learning workspace. He uses both built-in roles and customised roles to limit tasks users can perform and control access to resources.

Ensure accountability with Azure resources

Phil uses the quotas capability to configure resource limits and sharing. This helps him manage spending, control costs, and ensure equitable resource allocation across teams. It also encourages accountability by the respective project teams.






Achieve model governance and compliance

Achieve model governance and compliance

Phil can now see what’s in production and ensure organisational compliance through auditing, role-based access control, and quota management. He appreciates how Azure Machine Learning not only supports the data science and operations teams, but also delivers the enterprise-grade governance capabilities he needs.

Azure Machine Learning provides the organisational controls essential for making machine learning projects successful and more secure.


Monitoring Azure ML
Plan and manage costs for Azure ML

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 is in production, as well as ensure organisational compliance through auditing, role-based access control and quota management.

Azure ML provides the organisational controls essential for making machine learning projects successful and secure.


Monitoring Azure ML
Plan and manage costs for Azure ML









See how customers are delivering value with MLOps









“With MLOps in Machine Learning, we’ve improved bus departure predictions by 74 percent, and riders spend 50 percent less time waiting.”

Sze-Wan Ng: Director of Analytics & Development, TransLink


  • Read the story


TransLink




“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.”

Vijaya Sekhar Chennupati, Applied data scientist, Johnson Controls

Johnson Controls


























Related Posts

MLOps: Machine learning operations | Algorithmia

MLOps: Machine learning operations | Algorithmia Nothing gets to production without MLOps. Here’s everything you need to know about MLOps—and why it’s the key to generating business value…

Machine learning operations – Cloud Adoption Framework

Machine learning operations – Cloud Adoption Framework Article 08/30/2022 18 minutes to read …

What is MLOps? – Databricks

What is MLOps? – Databricks Back to glossary What is MLOps? MLOps stands for Machine Learning Operations. MLOps is a core function of Machine Learning engineering, focused on…

MLOps: Machine learning model management – Azure Machine Learning

MLOps: Machine learning model management – Azure Machine Learning Article 08/26/2022 9 minutes to read …

Machine Learning Operations – MLOps | Microsoft Azure

Machine Learning Operations – MLOps | Microsoft Azure …

Machine Learning for Operations

Machine Learning for Operations Managing infrastructure is a complex problem with a massive amount of signals and many actions that can be taken in response; that’s the classic…

Leave a Reply