Carl Osipov - MLOps Engineering at Scale

A useful piece of feedback that I got from a reviewer of this book was that it became a cheat code for them to scale the steep MLOps learning curve. I hope that the content of this book will help you become a better informed practitioner of machine learning engineering and data science, as well as a more productive contributor to your projects, your team, and your organization.

In 2021, major technology companies are vocal about their efforts to democratize artificial intelligence (AI) by making technologies like deep learning more accessible to a broader population of scientists and engineers. Regrettably, the democratization approach taken by the corporations focuses too much on core technologies and not enough on the practice of delivering AI systems to end users. As a result, machine learning (ML) engineers and data scientists are well prepared to create experimental, proof-of-concept AI prototypes but fall short in successfully delivering these prototypes to production. This is evident from a wide spectrum of issues: from unacceptably high failure rates of AI projects to ethical controversies about AI systems that make it to end users. I believe that, to become successful, the effort to democratize AI must progress beyond the myopic focus on core, enabling technologies like Keras, PyTorch, and TensorFlow. MLOps emerged as a unifying term for the practice of taking experimental ML code and running it effectively in production. Serverless ML is the leading cloud-native software development model for ML and MLOps, abstracting away infrastructure and improving productivity of the practitioners.

I also encourage you to make use of the Jupyter notebooks that accompany this book. The DC taxi fare project used in the notebook code is designed to give you the practice you need to grow as a practitioner. Happy reading and happy coding!

I am forever grateful to my daughter, Sophia. You are my eternal source of happiness and inspiration. My wife, Alla, was boundlessly patient with me while I wrote my first book. You were always there to support me and to cheer me along. To my father, Mikhael, I wouldnt be who I am without you.

I also want to thank the people at Manning who made this book possible: Marina Michaels, my development editor; Frances Buontempo, my technical development editor; Karsten Strbaek, my technical proofreader; Deirdre Hiam, my project editor; Michele Mitchell, my copyeditor; and Keri Hales, my proofreader.

Many thanks go to the technical peer reviewers: Conor Redmond, Daniela Zapata, Dianshuang Wu, Dimitris Papadopoulos, Dinesh Ghanta, Dr. Irfan Ullah, Girish Ahankari, Jeff Hajewski, Jess A. Jurez-Guerrero, Trichy Venkataraman Krishnamurthy, Lucian-Paul Torje, Manish Jain, Mario Solomou, Mathijs Affourtit, Michael Jensen, Michael Wright, Pethuru Raj Chelliah, Philip Kirkbride, Rahul Jain, Richard Vaughan, Sayak Paul, Sergio Govoni, Srinivas Aluvala, Tiklu Ganguly, and Todd Cook. Your suggestions helped make this a better book.

Thank you for purchasing MLOps Engineering at Scale.

To get the most value from this book, youll want to have existing skills in data analysis with Python and SQL, as well as have some experience with machine learning. I expect that if you are reading this book, you are interested in developing your expertise as a machine learning engineer, and you are planning to deploy your machine learningbased prototypes to production.

This book is for information technology professionals or those in academia who have had some exposure to machine learning and are working on or are interested in launching a machine learning system in production. There is a refresher on machine learning prerequisites for this book in appendix A. Keep in mind that if you are brand new to machine learning you may find that studying both machine learning and cloud-based infrastructure for machine learning at the same time can be overwhelming.

If you are a software or a data engineer, and you are planning on starting a machine learning project, this book can help you gain a deeper understanding of the machine learning project life cycle. You will see that although the practice of machine learning depends on traditional information technologies (i.e., computing, storage, and networking), it is different from the traditional information technology in practice. The former is significantly more experimental and more iterative than you may have experienced as a software or a data professional, and you should be prepared for the outcomes to be less known in advance. When working with data, the machine learning practice is more like the scientific process, including forming hypotheses about data, testing alternative models to answer questions about the hypothesis, and ranking and choosing the best performing models to launch atop your machine learning platform.

If you are a machine learning engineer or practitioner, or a data scientist, keep in mind that this book is not about making you a better researcher. The book is not written to educate you about the frontiers of science in machine learning. This book also will not attempt to reteach you the machine learning basics, although you may find the material in appendix A, targeted at information technology professionals, a useful reference. Instead, you should expect to use this book to become a more valuable collaborator on your machine learning team. The book will help you do more with what you already know about data science and machine learning so that you can deliver ready-to-use contributions to your project or your organization. For example, you will learn how to implement your insights about improving machine learning model accuracy and turn them into production-ready capabilities.