Implementing MLOps in the Enterprise

Chapter 5. Developing Models for Production

Developing ML models is no longer confined to experimental labs and research papers. It’s about real-world applications, and that means production. That’s why building high-performing models is at the heart of developing models for production.

A production-first mindset ensures that the models actually make it to production and answer real-life business cases. Otherwise, models get stuck throughout the ML pipeline due to lack of collaboration between teams, technological discrepancies, or other types of friction.

This chapter focuses on building the best models you can. It details all the steps and processes to implement and run on models throughout the ML pipeline before production. This includes running, tracking, and comparing ML jobs, automations, training and ML at scale; testing; resource management; and much more. It details various methodologies, tools, and approaches, together with code examples you can follow.

When following the steps and trying out the exercises at the end of the chapter, be conscious of the entire MLOps pipeline and how your work could be integrated and automated together with the other steps you or other team members are taking. By taking these steps with a production-first approach in mind, you can assure the reliability, stability, and performance of your ML models.

AutoML

Building the best ML model is an iterative process that relies on data science experience and intuition. The data scientist attempts ...