MLOps tries to create continuously deployable workflows, and that requires tests, checks and scripts chained that will be run in a CI/CD environment. We will learn how to work with these, focusing in the most popular tool nowadays, GitHub Actions.
The student will learn different concepts associated with CI/CD workflows, and will be able to create their own using GitHub actions.
The student must be familiar with the concepts involving CI/CD and how to use readily available ones for data science/machine learning and will have used them in the data science project that’s been developed for this course.
This step is related to all the stages that have continuous in their name in the MLOps ladder: continuous training means that, as new data comes in, it needs to be either fed into the model for training or simply passed through the ETL pipeline to check for health and eventually represent it in some sensible way.
In most cases, cloud providers and software as a service (SaaS) companies will try to offer you a full-span solution for MLOps, where you will have to pay every step of the way. It will also be a monolithic solution where the onramp to start and using it from scratch, as well as stopping its use or moving it to other tool, can be extremely complicated or even impossible.
In general, however, creating MLOps workflows can be done easily and for free if you are working with free software and your repositories have been released with a free license.
MLOps, in general, is strongly related to the movement for public money, public software, as well as open science. So working in public repositories is the right thing to do if you are in Academia; it is probably a very good idea if you are in industry and you are not working with proprietary data sets.
Come this point, different parts of the machine learning pipeline, like downloading and processing data, must be already available. A series of workflows using GitHub actions must be created for
Different workflows must be triggered by code and by data changes, and changes in the repository should be made automatically by these workflows without human intervention.