Chapter 16: Bringing Models into Production with MLOps
In the previous chapter, we looked into model interoperability using ONNX, hardware optimization using FPGAs, and the integration of trained models into other services and platforms. So far, you have learned how to implement each step in an end-to-end machine learning pipeline with data cleansing, preprocessing, labeling, experimentation, model training, optimization, and deployment. In this chapter, we will connect the bits and pieces from all the previous chapters to integrate and automate them in a build and release pipeline. We will reuse all these concepts to build a version-controlled, reproducible, automated ML training and deployment process as a continuous integration and continuous deployment (CI/CD) pipeline in Azure. In analogy to the DevOps methodology in software development, we will refer to this topic as MLOps in ML.
First, we will take a look at how to produce reproducible builds, environments, and deployments...