Deploying Stochastic Systems

Abstract

If you’ve made it this far, you already have the skills to build a complete end to end data science system. Data science of course is more than machine learning and code which are really only tools, and to build end to end systems, we need to understand people, processes, and technology, so this chapter will take a step back and give you a bird’s-eye view of the entire MLOps lifecycle, tying in what we’ve learned in previous chapters to formally define each stage. Once we have the lifecycle defined, we’ll be able to analyze it to understand how we can reduce technical debt by considering the interactions between the various stages from data collection and data engineering through to model development and deployment. We’ll cover some philosophical debates between model-centric vs. data-centric approaches to MLOps and look at how we can move toward continuous delivery, the ultimate litmus test for how much value your models are creating in production. We will also discuss how the rise of generative AI may impact data science development in general, build a CI/CD pipeline for our toolkit, and talk about how we can use pre-build cloud services to deploy your models. Without further ado, let’s explore the stages of the ML lifecycle again and introduce the spiral ML lifecycle formally.