General Overview of Machine Learning Operations (MLOps)
Principles, Components, Roles and Architecture Best Practices of a successful MLOps project
During the last couple of years, I’ve been implementing Machine Learning systems in real production environments for clients from Consumer Packaged Goods (CPG) and Healthcare.
While understanding business requirements, designing the model and interacting with raw data are still huge challenges, one of the most interesting and exciting part was the project industrialization.
One of the issues we had, and still have, is to understand what are the best practices in the industry we should adopt when maintaining a machine learning system in production. A recent work from Kreuzberger et al. [1] provides an overview of what Machine Learning Operations (MLOps) are, what are its principles, what architecture components enable those principles, what are the roles in an MLOps project and what can be an architecture of such a system. I think it also sets the best practices from an academic perspective on the topic.
Principles of MLOps
To better understand what MLOps is and what it should achieve, they first identify some principles such a project should respect: