Mastering MLOps Architecture From Code to Deployment Manage the production cycle of continual learning ML models with MLOps… (Raman Jhajj) (z-library.sk, 1lib.sk, z-lib.sk)

Mastering MLOps Architecture: From Code to Deployment Manage the production cycle of continual learning ML models with MLOps Raman Jhajj

www.bpbonline.com

Copyright © 2024 BPB Online 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 BPB Online 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. BPB Online has endeavored to provide trademark information about all of the companies and products mentioned in this book by the appropriate use of capitals. However, BPB Online cannot guarantee the accuracy of this information. First published: 2024 Published by BPB Online WeWork 119 Marylebone Road London NW1 5PU UK | UAE | INDIA | SINGAPORE ISBN 978-93-55519-498 www.bpbonline.com

Dedicated to My family, that gave me the gift of dreams and Friends, who became family.

About the Author Raman Jhajj is a passionate leader in the data and software engineering space with experience building high-performing teams and leading organizations to become data-driven. He has experience in leading the development of SaaS applications, modern data platforms and MLOps infrastructure. He brings technical expertise across the data stack including AWS, Python, Django, Java, PostgreSQL, Hadoop, Spark, Kafka, Docker, CI/CD, SQL, NoSQL, and more. Raman holds a master’s degree in applied computer science from Georg- August University, Germany as well as a bachelor’s in computer science from ICFAI University, India. After living in India, Germany, Austria, and Malta, he now calls Canada home. Over the course of his career, Raman has driven key initiatives around modernizing data infrastructure, establishing data engineering capabilities, and building MLOps platforms. Raman thrives on bringing cross-functional teams together to ensure alignment between technology and business goals. He has a proven track record of mentoring engineers and nurturing their potential. When he is not working, you can often find him reading, writing, or exploring new places and cultures. He is passionate about using technology for social good, driven by a mission to leverage data engineering and AI for positive change.

About the Reviewer Ashish Patel, an accomplished author, data scientist and researcher with over 11 years of experience. He is a luminary in predictive modeling, data preprocessing, feature engineering, machine learning, and deep learning. Notably, Ashish has taken center stage as a keynote speaker at prestigious events like AWS Community Day, AWS AI ML Days, Faculty Development Programs (FDPs), and IIT Techfest, captivating audiences with his insights. Currently serving as the Sr. AWS AI ML Solution Architect at IBM India Pvt Ltd, he architects innovation by collaborating with IBM and AWS specialists to craft enterprise solutions on Red Hat OpenShift, AWS Infrastructure, and IBM Software technology, aligning seamlessly with the AWS Well-Architected Framework. Ashish is a five- time LinkedIn Top Voice and an AI Research Scientist, with expertise spanning MLOps and a multitude of LLMs and FM Models. Recognized on LinkedIn for his contributions in Statistics, Data Science, Data Analytics, AI, and Machine Learning, Ashish is also a GitHub sensation with over 5k+ followers, marking his profound impact in open-source communities. In the realm where data reigns supreme, Ashish Patel crafts, speaks, and influences the future. He a Quantum Machine Learning practitioner and researcher working with international research community.

Acknowledgement Writing a book is harder than I thought and more rewarding than I could have ever imagined. None of this would have been possible without the support of my family and friends, whom I would like to acknowledge and thank. I would like to start by thanking my awesome wife, Simran for being the constant support from those late-night writing sessions and frustration-filled days to my ramblings of how hard it is to put thoughts on paper. I want to thank my parents - Dad for constantly guiding and showing me that writing a book is an achievable target and Mom for her unwavering belief in me. I thank Kanwar, Kuljeet and Garima for their constant support throughout the ups and downs of life and for always being there for me. I thank Kaisha for those video calls and for filling the days with laughter. To all those friends who have been a part of my getting here: Parminder, Kiran, Anmol, Harman, Jagvir and Sukhpreet, I thank you for your heartfelt support and ready smiles, shared meals, advice, perspectives, and friendships. I thank Baani and Ravtaj for the playtime and for reminding me of what it is like to be a child again. To my mentors throughout this journey: Malaika, Dean Chen, Michiah, and Tovah, I thank you for being the leaders I trust, honour, and respect. To everyone at BPB Publications who enabled me to write this book. Thank you for the guidance and expertise in bringing this book to fruition. It was a long journey of revising this book, with valuable participation and collaboration of reviewers, technical experts, and editors. I would also like to acknowledge the valuable contributions of my colleagues and co-workers who have taught me so much and provided

valuable feedback on my work, during many years working in the tech industry. Finally, I want to thank you, my cherished readers, for taking an interest in my book. To have it received by you is an unexpected gift that keeps me grounded in the moment.

Preface MLOps is the intersection of DevOps, data engineering and machine learning. Working in the field of machine learning is highly dependent on ever-changing data, whereas MLOps is needed to deliver excellent ML and AI results. This book provides a practical guide to MLOps for data scientists, data engineers, and other professionals involved in building and deploying machine learning systems. It introduces MLOps, explaining its core concepts like continuous integration and delivery for machine learning. It outlines MLOps components and architecture, providing an understanding of how MLOps supports robust ML systems that continuously improve. By covering the end-to-end machine learning pipeline from data to deployment, the book helps readers implement MLOps workflows. It discusses techniques like feature engineering, model development, A/B testing, and canary deployments. The book equips readers with knowledge of MLOps tools and infrastructure for tasks like model tracking, model governance, metadata management, and pipeline orchestration. Monitoring and maintenance processes to detect model degradation are covered in depth. With its comprehensive coverage and practical focus, this book enables data scientists, data engineers, DevOps engineers, and technical leaders to effectively leverage MLOps. Readers can gain skills to build efficient CI/CD pipelines, deploy models faster, and make their ML systems more reliable and production-ready. Overall, the book is an indispensable guide to MLOps and its applications for delivering business value through continuous machine learning and AI. Chapter 1: Getting Started with MLOps - This chapter introduces MLOps, explaining how it combines machine learning, DevOps, and data engineering to enable continuous delivery of ML models. It covers the importance of MLOps, its principles like reproducibility and auditability,

best practices, and strategies for implementation. The difference between MLOps and the traditional software engineering and the unique challenges of productionizing machine learning are also discussed. The chapter provides a foundation for understanding the MLOps methodology. Chapter 2: MLOps Architecture and Components - This chapter covers the architecture and components of MLOps systems. It discusses the building blocks like data pipelines, model training, deployment, monitoring, and orchestration. The chapter outlines reference architectures for different maturity levels, from basic to enterprise-grade. It explains environment semantics and model deployment patterns. Finally, it walks through an end- to-end workflow integrating all components across development, staging, and production environments. The goal is to provide a foundation for designing and implementing MLOps solutions suitable for various use cases. Chapter 3: MLOps Infrastructure and Tools - This chapter explores the infrastructure and tools needed for MLOps. It covers key components like storage, compute, containers, orchestration platforms, and ML platforms for deployment, model registries, and feature stores. The chapter discusses public cloud versus on-premises options, standardized development environments, and build versus buy decisions. It aims to provide guidance on setting up a robust, scalable infrastructure tailored to an organization’s specific use cases and resources. Chapter 4: What are Machine Learning Systems? - This chapter explains what machine learning systems are and how they differ from ML research. It covers an implementation roadmap with phases for initial development, transition to operations, and ongoing operations. The chapter discusses using standardized project structures like cookiecutter data science to facilitate eventual productionization. It aims to provide a foundation for taking a full systems approach to developing real-world ML applications, not just algorithms. The goal is to equip readers with an understanding of all components needed to build successful ML systems. Chapter 5: Data Preparation and Model Development - This chapter covers data preparation and model development within the MLOps lifecycle. It discusses best practices for version control, preparing data, performing exploratory analysis, feature engineering, training models, and

tracking experiments with MLflow. The chapter shows how these steps fit into a standardized project structure to enable collaboration and reproducibility. It aims to provide guidance on implementing key phases of the machine learning lifecycle in a way that facilitates eventual operationalization and automation. Chapter 6: Model Deployment and Serving - This chapter covers model deployment and serving in the MLOps lifecycle. It explores strategies like static, dynamic, and streaming deployment, comparing deployment on devices versus servers using VMs, containers, or serverless technologies. The chapter discusses inference options like batch processing versus real- time APIs. It also looks at deployment patterns like canary releases and multi-armed bandits for controlled model rollout. Chapter 7: Continuous Delivery of Machine Learning Models - This chapter explores methods for implementing continuous integration, continuous training, and continuous delivery in machine learning systems. It examines ML/AI pipelines and architectural maturity levels. Key topics include continuous integration tools like GitHub Actions, strategies for determining when and what to retrain models on, and considerations for rapidly deploying updated models into production through continuous delivery. Chapter 8: Continual Learning - This chapter explores continual learning in machine learning systems, which involves models perpetually learning and adapting to new data without forgetting past knowledge. It covers principles like stateful training, challenges around obtaining fresh data and evaluating updates, and implementing continual learning in MLOps through triggers and robust monitoring. The goal is to enable frequent automated model updates while maintaining safety, transparency and control. Chapter 9: Continuous Monitoring, Logging, and Maintenance - This chapter covers principles and best practices for monitoring machine learning models across environments. It examines why continuous monitoring matters, integrating it into MLOps workflows, logging model metadata and performance data, using frameworks like Evidently and Alibi Detect, and evaluating models with techniques like A/B testing.

Code Bundle and Coloured Images Please follow the link to download the Code Bundle and the Coloured Images of the book: https://rebrand.ly/mn9abap The code bundle for the book is also hosted on GitHub at https://github.com/bpbpublications/Mastering-MLOps-Architecture- From-Code-to-Deployment. In case there's an update to the code, it will be updated on the existing GitHub repository. We have code bundles from our rich catalogue of books and videos available at https://github.com/bpbpublications. Check them out! Errata We take immense pride in our work at BPB Publications and follow best practices to ensure the accuracy of our content to provide with an indulging reading experience to our subscribers. Our readers are our mirrors, and we use their inputs to reflect and improve upon human errors, if any, that may have occurred during the publishing processes involved. To let us maintain the quality and help us reach out to any readers who might be having difficulties due to any unforeseen errors, please write to us at : errata@bpbonline.com Your support, suggestions and feedbacks are highly appreciated by the BPB Publications’ Family. Did you know that BPB offers eBook versions of every book published, with PDF and ePub files available? You can upgrade to the eBook version at www.bpbonline.com and as a print book customer, you are entitled to a discount on the eBook copy. Get in touch with us at : business@bpbonline.com for more details.

At www.bpbonline.com, you can also read a collection of free technical articles, sign up for a range of free newsletters, and receive exclusive discounts and offers on BPB books and eBooks. Piracy If you come across any illegal copies of our works in any form on the internet, we would be grateful if you would provide us with the location address or website name. Please contact us at business@bpbonline.com with a link to the material. If you are interested in becoming an author If there is a topic that you have expertise in, and you are interested in either writing or contributing to a book, please visit www.bpbonline.com. We have worked with thousands of developers and tech professionals, just like you, to help them share their insights with the global tech community. You can make a general application, apply for a specific hot topic that we are recruiting an author for, or submit your own idea. Reviews Please leave a review. Once you have read and used this book, why not leave a review on the site that you purchased it from? Potential readers can then see and use your unbiased opinion to make purchase decisions. We at BPB can understand what you think about our products, and our authors can see your feedback on their book. Thank you! For more information about BPB, please visit www.bpbonline.com. Join our book’s Discord space Join the book’s Discord Workspace for Latest updates, Offers, Tech happenings around the world, New Release and Sessions with the Authors: https://discord.bpbonline.com

Table of Contents 1. Getting Started with MLOps Introduction Structure Objectives Understanding MLOps Experimentation and tracking Model management Importance of MLOps The evolution of MLOps Software engineering projects versus machine learning projects DevOps versus MLOps Principles of MLOps MLOps best practices Code Data Model Metrics and KPIs Deployment Team MLOps in an organization MLOps strategy Cloud Training and talent

Vendor Executive focus on Return on Investment Implementing MLOps Overcoming challenges of MLOps MLOps in Cloud MLOps on-premises MLOps in hybrid environments Conclusion Points to remember Key terms 2. MLOps Architecture and Components Introduction Structure Objectives MLOps components Data source and data versioning Data analysis and experiment management Code repository Pipeline orchestration Workflow orchestration CI/CD automation Model training and storage Model training Model registry Model deployment and serving Monitoring for model, data, and application Training performance tracking Metadata store Feature processing and storage Feature processing

Feature store MLOps architecture Architecture level 1: Minimum viable architecture Architecture level 2: Production grade MLOps Architecture level 3: Enterprise grade MLOps The semantics of dev, staging, and production Execution environment Code Models Data Machine learning deployment patterns Deploy models Deploy code Bringing the architectural components together Development environment Staging environment Production environment Conclusion Points to remember Key terms 3. MLOps Infrastructure and Tools Introduction Structure Objectives Getting started with infrastructure Storage Extract, transform, load/extract, load, transform Batch processing and stream processing Compute

Public Cloud vendors versus private data centers Development environments Development environment setup Integrated development environments Containers Orchestration/workflow management Airflow installation Installing using PyPi Installing in Docker Airflow in production Example: Airflow Direct Acyclic Graphs Machine learning platforms Model deployment Model registry Feature store Installing MLflow Build versus buy Conclusion Points to remember Key terms 4. What are Machine Learning Systems? Introduction Structure Objectives What is a machine learning system Machine learning systems use cases Understanding machine learning systems Machine learning in research versus production Objectives and requirements

Computational priorities Data Fairness Interpretability An implementation roadmap for MLOps-based machine learning systems Phase 1: Initial development Phase 2: Transition to operations Phase 3: Operations Machine learning development: Cookiecutter data science project structure What is cookiecutter Why cookiecutter Getting started with cookiecutter data science Repository structure Conclusion Points to remember Key terms 5. Data Preparation and Model Development Introduction Structure Objectives MLOps code repository best practices pre-commit hooks Data sourcing Data sources Data versioning Exploratory data analysis Data preparation