MLOps: Say hello to Continuous Training (CT)

How to expand the CI/CD concepts to Machine Learning projects by adding Continous Training (CT) in your MLOps pipelines (a real example).

If you ever put a Machine Learning model in production, you probably have wondered how can I automatically integrate new versions of the model with either a new model’s architecture or fresh new data? Although it is a process simple to describe, it represents an important challenge to automate.

We have been told that MLOps, the practice that advocates for automation and monitoring of all steps in the ML model’s construction, is the answer to such problems (among others). We see people talking about this everywhere. However, somehow the question remains being asked. Is it because it is cool to talk about MLOps or maybe because we haven’t cracked the idea yet?

Considering that MLOps is an extension of the DevOps practice, it makes sense for it to build on top of the modern agile development practice.

It is important to notice from the previous chart that you can’t have DevOps (or MLOps) without a repository for source control (git). It’s the fundamental piece that connects the dev with the ops. We have already touched base on the importance of the git repository in…