16 Production infrastructure
This chapter covers
- Using and implementing a model registry in passive retraining paradigms to ensure ML solutions adapt to drift.
- Using a feature store to eliminate rework and share calculated insights throughout an organization.
- Selecting appropriate serving layer architecture to minimize both complexity and cost for a particular use case.
Utilizing ML in a real-world use case to solve a complex problem is challenging. The sheer number of skills needed to take a company’s data (frequently messy, partially complete, and rife with quality issues), select an appropriate algorithm, tune a pipeline, and validate that the prediction output of a model (or an ensemble of models) solves the problem to the satisfaction of the business is quite daunting. The complexity of an ML-backed project does not end with the creation of an acceptably performing model, though. The architectural considerations, choices to be made on such, and the implementation details of which can add significant challenges to a project if they aren’t made correctly.
Every day there seems to be a new open-sourced tech stack that promises an easier deployment strategy or a magical automated solution that meets the needs of all. With this deluge of tools and platforms constantly releasing, it is daunting to know where to go to meet the needs of a particular project.