Understanding the MLOps implementation
As mentioned, MLOps is a concept, not an implementation. We will provide an implementation of MLOps as the foundation for this chapter. We will establish an Azure DevOps pipeline to orchestrate an AML pipeline for transforming data in the dev environment, create a model leveraging MLflow, and evaluate whether the model is performing better or equal to the existing model. Following this pipeline, if a new model is registered, we will deploy this new model in the dev environment leveraging blue/green deployments. Blue/green deployments enable high availability. As a new model is deployed, the existing endpoint deployment is available during the new model deployment. After the new model is deployed to the managed online endpoint, we swap traffic over to the new model. After the model is deployed in the dev environment, we will trigger an approval process to then register and deploy the new model into the qa environment, which will leverage blue/green...
