Trenton Potgieter - Automated Machine Learning on AWS: Fast-track the development of your production-ready machine learning applications the AWS way

Fast-track the development of your production-ready machine learning applications the AWS way

Trenton Potgieter

BIRMINGHAMMUMBAI

Copyright 2022 Packt Publishing

All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, without the prior written permission of the publisher, except in the case of brief quotations embedded in critical articles or reviews.

Every effort has been made in the preparation of this book to ensure the accuracy of the information presented. However, the information contained in this book is sold without warranty, either express or implied. Neither the author, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to have been caused directly or indirectly by this book.

Packt Publishing has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy of this information.

Publishing Product Manager: Devika Battike

Senior Editor: Nathanya Dias

Content Development Editor: Nazia Shaikh

Technical Editor: Devanshi Ayare

Copy Editor: Safis Editing

Project Coordinator: Aparna Ravikumar Nair

Proofreader: Safis Editing

Indexer: Sejal Dsilva

Production Designer: Roshan Kawale

Marketing Coordinator: Abeer Dawe, Shifa Ansari

First published: April 2022

Production reference: 1100322

Published by Packt Publishing Ltd.

Livery Place

35 Livery Street

Birmingham

B3 2PB, UK.

ISBN 978-1-80181-182-8

www.packt.com

Virtually everyone struggles with operationalizing machine learning models. Training your first model can sometimes seem like an insurmountable challenge, until you realize that you also need an end-to-end pipeline to supply new data for inference and retraining the model when its performance inevitably degrades. Although AWS offers the broadest and deepest set of machine learning services, figuring out where to get started and how to tie all those options together normally requires months of painful experimentation. This book cuts through the uncertainty based on Trenton's first-hand experiences working with both the most sophisticated technology companies in the world as well as organizations new to machine learning.

I've worked with hundreds of companies around the world trying to get value from artificial intelligence and machine learning. The problem is that machine learning can mean very different things even within the same company, much less across different organizations or industries. Some teams are just starting to invest in AI and machine learning and want to build their first model, while other teams in the same organization want to scale up sophisticated experimentation and monitoring frameworks to support thousands of models in production. Most companies hire data scientists or machine learning engineers with skill mismatches in the hope that they'll figure it out. Trenton has the rare advantage of seeing how large organizations have successfully scaled up their modeling pipelines as well as where they've faltered. Even more importantly, he has hard-won experience helping them solve those challenges.

The machine learning space evolves so quickly that focusing on any single algorithm, package, or platform can lead to outdated content. Trenton avoids this trap by translating timeless software engineering concepts like continuous integration and continuous delivery to the machine learning space. Unlike many approaches, however, he punctuates each concept with hands-on examples to illustrate how everything works in practice so that you don't need to struggle to translate theory to real life applications.

For example, data scientists often view automated machine learning with disdain due to previous exposure to automation that felt more like a straitjacket than an accelerant. People new to machine learning as well as sophisticated data scientist can overlook AutoML on AWS due to inexperience or ignorance of its benefits. Understanding when and why to use AutoML to get an initial benchmark on a new project or avoid manually selecting and tuning algorithms every time you retrain a model can reduce the time you spend on model training by an order of magnitude.

Even more importantly, learning how to think about the long-term maintenance of the machine learning pipelines will help you avoid painful decisions on whether to spend time refactoring existing models or deliver new projects. Software engineers have been leveraging CI/CD processes for over a decade at this point, but most machine learning practitioners aren't aware of best practices from the DevOps space. Most data scientists discover the need for this process only after they've built a few models and realized that reusable model assets and pipelines are required if they want to do anything beyond maintaining brittle modeling workflows by hand.

Finally, Trenton highlights concepts like source-code and data-centric machine learning that normally require hiring working at a top technology company that's overcome scaling challenges that most companies don't experience early on in their machine learning journeys. Most people and organizations hit a wall after implanting a CI/CD pipeline and building their first. They run up against the challenges of scheduling, tracking, and monitoring their machine learning pipelines. This book is the only example I'm aware of that offers prescriptive guidance on how to structure long-term machine learning pipelines and avoid the common pitfalls that machine learning teams typically encounter.

In short, the concepts in this book will help you move beyond the hopes and dreams of machine learning, to getting machine learning applications into production and delivering value.

Jonathan Dahlberg

Head of ML Solution Engineering

Snorkel AI

Trenton Potgieter is a senior AI/ML specialist at AWS and has been working in the field of ML since 2011. At AWS, he assists multiple AWS customers to create ML solutions and has contributed to various use cases, broadly spanning computer vision, knowledge graphs, and ML automation using MLOps methodologies. Trenton plays a key role in evangelizing the AWS ML services and shares best practices through forums such as AWS blogs, whitepapers, reference architectures, and public-speaking events. He has also actively been involved in leading, developing, and supporting an internal AWS community of MLOps-related subject matter experts.

Hemanth Boinpally is a Machine Learning Engineer at AWS. He has several years of experience working in data science and ML. He has worked with enterprise customers across different industries, such as healthcare, finance, logistics, and manufacturing. He enjoys providing end-to-end ML solutions for complex business problems. His expertise across the technology stack helps him collaborate with cross-functional teams to build successful ML products. This includes engaging with business stakeholders, the research and development of ML models, and operationalizing these models using MLOps principles. He has worked in areas such as model bias detection, interpretable models, NLP, CV, active learning, and deep learning.