[WIP] Real-Time Machine Learning: A Practitioner’s Guide
With Examples from Fintech
Hello!
Welcome to Real-Time Machine Learning: A Practitioner’s Guide. This book provides practical guidance for building and maintaining real-time machine learning systems in production as an engineering discipline.
About This Book
This book covers the entire lifecycle of real-time ML systems, from initial project planning to steady-state operation of a single model to scaling to multiple models across an organization.
The scope of this book is ML systems in reasonable-scale organizations.
“Reasonable-scale” organizations are the sweet spot for this book, because (i) they have reached the level where there’s data and real use-cases for real-time ML and (ii) the challenges they face are similar to those faced by other companies, so here is where an example-led approach works best.
Examples from Fintech
I firmly believe that the best way to teach is by leading with simple examples to quickly convey the big picture and then add an explanation, following up with more advanced examples.
Financial organizations such as banks and insurance companies have been using applied ML for a long time. There are interesting and complex problems, which allow us to explore the many dimensions and trade-offs involved in real-time ML. These include: adversarial incentives, regulatory constraints, market pressure from competition, cost concerns, high cost of failure, explainability requirements, among many more.
Fintech is just the modern iteration of traditional financial organizations so it’s a natural place to draw examples from.
Patterns and Lessons Learned
In addition to examples, patterns and lessons learned are two ways I found to convey experience from operating real-time ML systems. These come from my own learnings and from the literature at large.
Patterns: things that are self-contained and common enough to merit a short “name”. Examples: shadow-mode deployment, pre-mortem, walking skeleton.
Lessons Learned: These are short sentences with statements of fact, suggestions or warnings about a particular aspect of real-time ML.
What this book is not
This is not an introductory book to ML. It assumes the reader already understands what ML is and how it’s used.
This book is not aimed at teaching you how to use a particular tool or framework. It is focused on high-level concepts and ideas instead.
Is there a PDF version of the book?
Yes. Every new release triggers a new PDF version. Download the latest PDF here.