You're reading from Engineering MLOps Rapidly build, test, and manage production-ready machine learning life cycles at scale

Product type Paperback

Published in Apr 2021

Publisher Packt

ISBN-13 9781800562882

Length 370 pages

Edition 1st Edition

Tools

Azure Functions

Concepts

Machine Learning

Author (1):

Emmanuel Raj

View More author details

Table of Contents (18) Chapters

Preface

1. Section 1: Framework for Building Machine Learning Models

2. Chapter 1: Fundamentals of an MLOps Workflow FREE CHAPTER

3. Chapter 2: Characterizing Your Machine Learning Problem

4. Chapter 3: Code Meets Data

5. Chapter 4: Machine Learning Pipelines

6. Chapter 5: Model Evaluation and Packaging

7. Section 2: Deploying Machine Learning Models at Scale

8. Chapter 6: Key Principles for Deploying Your ML System

9. Chapter 7: Building Robust CI/CD Pipelines

10. Chapter 8: APIs and Microservice Management

11. Chapter 9: Testing and Securing Your ML Solution

12. Chapter 10: Essentials of Production Release

13. Section 3: Monitoring Machine Learning Models in Production

14. Chapter 11: Key Principles for Monitoring Your ML System

15. Chapter 12: Model Serving and Monitoring

16. Chapter 13: Governing the ML System for Continual Learning

17. Other Books You May Enjoy

Why package ML models?

MLOps enables a systematic approach to train and evaluate models. After models are trained and evaluated, the next steps are to bring them to production. As we know, ML doesn't work like traditional software engineering, which is deterministic in nature and where a piece of code or module is imported into the existing system and it works. Engineering ML solutions is non-deterministic and involves serving ML models to make predictions or analyze data.

In order to serve the models, they need to be packed into software artifacts to be shipped to the testing or production environments. Usually, these software artifacts are packaged into a file or a bunch of files or containers. This allows the software to be environment- and deployment-agnostic. ML models need to be packaged for the following reasons: