What this book covers
Chapter 1, Fundamentals of MLOps Workflow, gives an overview of the changing software development landscape by highlighting how traditional software development is changing to facilitate machine learning. We will highlight some daily problems within organizations with the traditional approach, showcasing why a change in thinking and implementation is needed. Proceeding that an introduction to the importance of systematic machine learning will be given, followed by some concepts of machine learning and DevOps and fusing them into MLOps. The chapter ends with a proposal for a generic workflow to approach almost any machine learning problem.
Chapter 2, Characterizing Your Machine Learning Problem, offers you a broad perspective on possible types of ML solutions for production. You will learn how to categorize solutions, create a roadmap for developing and deploying a solution, and procure the necessary data, tools, or infrastructure to get started with developing an ML solution taking a systematic approach.
Chapter 3, Code Meets Data, starts the implementation of our hands-on business use case of developing a machine learning solution. We discuss effective methods of source code management for machine learning, data processing for the business use case, and formulate a data governance strategy and pipeline for machine learning training and deployment.
Chapter 4, Machine Learning Pipelines, takes a deep dive into building machine learning pipelines for solutions. We look into key aspects of feature engineering, algorithm selection, hyperparameter optimization, and other aspects of a robust machine learning pipeline.
Chapter 5, Model Evaluation and Packaging, takes a deep dive into options for serializing and packaging machine learning models post-training to deploy them at runtime to facilitate machine learning inference, model interoperability, and end-to-end model traceability. You'll get a broad perspective on the options available and state-of-the-art developments to package and serve machine learning models to production for efficient, robust, and scalable services.
Chapter 6, Key Principles for Deploying Your ML System, introduces the concepts of continuous integration and deployment in production for various settings. You will learn how to choose the right options, tools, and infrastructure to facilitate the deployment of a machine learning solution. You will get insights into machine learning inference options and deployment targets, and get an introduction to CI/CD pipelines for machine learning.
Chapter 7, Building Robust CI and CD Pipelines, covers different CI/CD pipeline components such as triggers, releases, jobs, and so on. It will also equip you with knowledge on curating your own custom CI/CD pipelines for ML solutions. We will build a CI/CD pipeline for an ML solution for a business use case. The pipelines we build will be traceable end to end as they will serve as middleware for model deployment and monitoring.
Chapter 8, APIs and Microservice Management, goes into the principles of API and microservice design for ML inference. A learn by doing approach will be encouraged. We will go through a hands-on implementation of designing and developing an API and microservice for an ML model using tools such as FastAPI and Docker. You will learn key principles, challenges, and tips to designing a robust and scalable microservice and API for test and production environments.
Chapter 9, Testing and Securing Your ML Solution, introduces the core principles of performing tests in the test environment to test the robustness and scalability of the microservice or API we have previously developed. We will perform hands-on load testing for a deployed ML solution. This chapter provides a checklist of tests to be done before taking the microservice to production release.
Chapter 10, Essentials of Production Release, explains how to deploy ML services to production with a robust and scalable approach using the CI/CD pipelines designed earlier. We will focus on deploying, monitoring, and managing the service in production. Key learnings will be deployment in serverless and server environments using tools such as Python, Docker, and Kubernetes.
Chapter 11, Key Principles for Monitoring Your ML System, looks at key principles and aspects of monitoring ML systems in production for robust, secure, and scalable performance. As a key takeaway, readers will get a concrete explainable monitoring framework and checklist to set up and configure a monitoring framework for their ML solution in production.
Chapter 12, Model Serving and Monitoring, explains serving models to users and defining metrics for an ML solution, especially in the aspects of algorithm efficiency, accuracy, and production performance. We will deep dive into hands-on implementation and real-life examples on monitoring data drift, model drift, and application performance.
Chapter 13, Governing the ML System for Continual Learning, reflects on the need for continual learning in machine learning solutions. We will look into what is needed to successfully govern an ML system for business efficacy. Using the Explainable Monitoring framework, we will devise a strategy to govern and we will delve into the hands-on implementation for error handling and configuring alerts and actions. This chapter will equip you with critical skills to automate and govern your MLOps.