Sridhar Alla - Beginning MLOps with MLFlow : Deploy Models in AWS SageMaker, Google Cloud, and Microsoft Azure

Any source code or other supplementary material referenced by the author in this book is available to readers on GitHub via the books product page, located at www.apress.com/978-1-4842-6548-2 . For more detailed information, please visit www.apress.com/source-code .

This book is intended for all audiences ranging from beginners at machine learning, to advanced machine learning engineers, or even to machine learning researchers who wish to learn how to better organize their experiments.

The first two chapters cover the premise of the problem followed by the book, which is that of integrating MLOps principles into an anomaly detector model based on the credit card dataset. The third chapter covers what MLOps actually is, how it works, and why it can be useful.

The fourth chapter goes into detail about how you can implement and utilize MLFlow in your existing projects to reap the benefits of MLOps with just a few lines of code.

The fifth, sixth, and seventh chapters all go over how you can operationalize your model and deploy it on AWS, Microsoft Azure, and Google Cloud, respectively. The seventh chapter goes over how you can host a model on a virtual machine and connect to the server from an external source to make your predictions, so should any MLFlow functionality described in the book become outdated, you can always go for this approach and simply serve models on some cluster on the cloud.

The last chapter, Appendix, goes over how you can utilize Databricks, the creators of MLFlow, to organize your MLFlow experiments and deploy your models.

The goal of the book is to hopefully impart to you, the reader, knowledge of how you can use the power of MLFlow to easily integrate MLOps principles into your existing projects. Furthermore, we hope that you will become more familiar with how you can deploy your models to the cloud, allowing you to make model inferences anywhere on the planet so as long as you are able to connect to the cloud server hosting the model.

At the very least, we hope that more people do begin to adopt MLFlow and integrate it into their workflows, since even as a tool to organize your workspace, it massively improves the management of your machine learning experiments and allows you to keep track of the entire model history of a project.

Researchers may find MLFlow to be useful when conducting experiments, as it allows you to log plots on top of any custom-defined metric of your choosing. Prototyping becomes much easier, as you can now keep track of that one model which worked perfectly as a proof-of-concept and revert back to those same weights at any time while you keep tuning the hyperparameters. Hyperparameter tuning becomes much simpler and more organized, allowing you to run a complex script that searches over several different hyperparameters at once and log all of the results using MLFlow.

With all the benefits that MLFlow and the corresponding MLOps principles offer to machine learning enthusiasts of all professions, there really are no downsides to integrating it into current work environments. With that, we hope you enjoy the rest of the book!

Sridhar Alla

I would like to thank my wonderful wife, Rosie Sarkaria, and my beautiful, loving daughters, Evelyn and Madelyn, for all the love and patience during the many months I spent writing this book. I would also like to thank my parents, Ravi and Lakshmi Alla, for all the support and encouragement they continue to bestow upon me.

Suman Kalyan Adari

I would like to thank my parents, Venkata and Jyothi Adari, and my loving dog, Pinky, for supporting me throughout the entire process. I would especially like to thank my sister, Niharika Adari, for helping me with edits and proofreading and helping me write the appendix chapter.