MLOps

An ML workflow is a set of operations developed and executed to produce a mathematical model, which eventually is designed to solve a real-world problem. But these models have no value until they are deployed in production other than proofs of concept. ML models almost always require deployment to a production environment to provide business value.

At its core, MLOps fundamentally focuses on transitioning an experimental ML model into a fully operational production system. MLOps is an emerging practice, different from traditional DevOps due to the unique nature of the ML development life cycle and the specific ML artifacts it produces. The ML life cycle revolves around discerning patterns from training data, making the MLOps workflow particularly sensitive to changes in data, as well as variations in data volumes and quality.

A well-developed MLOps practice should support the monitoring of ML life cycle activities as well as the ongoing supervision of models once they...