MLOps in Practice
Shipping, Running and Improving Machine Learning Systems
- Pages
- 84
- Chapters
- 64
- Level
- Intermediate–Advanced
- Format
- E-book
About this book
Most of the work in a machine-learning system happens after the model is trained. This book covers that work: making models reproducible, deployable, observable and improvable.
You will learn to version data, code and models; build training pipelines; package and serve models; design deployment strategies such as shadow and canary releases; monitor data, predictions and outcomes; detect drift; retrain safely; and manage costs. Separate chapters address operations for language-model applications and the documentation and governance that regulators and auditors increasingly expect.
The book is tool-agnostic and grounded in real failure stories.
What you will learn
- Reproducibility: versioning data, code and models
- Training pipelines and experiment tracking
- Serving models: batch, online and streaming
- Safe deployment: shadow, canary and rollback
- Monitoring, drift detection and retraining
- Operating language-model applications and governance
Table of contents
- IntroductionIntroduction
- Chapter 1Why ML Systems Are Different
- Chapter 2The ML Life Cycle
- Chapter 3Hidden Technical Debt
- Chapter 4Failure Stories
- Chapter 5Maturity Levels
- Chapter 6Roles
- Chapter 7Defining Success Before Building
- Chapter 8Platform or Point Solutions?
- Chapter 9Reproducibility
- Chapter 10Versioning Code
- Chapter 11Versioning Data
- Chapter 12Versioning Models: The Registry
- Chapter 13Environments and Containers
- Chapter 14Training Pipelines
- Chapter 15Experiment Tracking
- Chapter 16Feature Engineering and Feature Stores
- Chapter 17Training–Serving Skew
- Chapter 18Testing ML Systems
- Chapter 19Packaging Models
- Chapter 20Batch Prediction
- Chapter 21Online Serving
- Chapter 22Streaming
- Chapter 23Edge Deployment
- Chapter 24Release Strategies
- Chapter 25Rollback
- Chapter 26Release Gates
- Chapter 27Scaling and Performance
- Chapter 28Reliability Engineering
- Chapter 29Security
- Chapter 30What to Monitor
- Chapter 31Data Drift
- Chapter 32Concept Drift
- Chapter 33Delayed Labels
- Chapter 34Alerting
- Chapter 35Retraining Strategies
- Chapter 36Champion and Challenger
- Chapter 37Feedback Loops
- Chapter 38Incident Response
- Chapter 39Cost Management
- Chapter 40Documentation
- Chapter 41Governance and Regulation
- Chapter 42Model Inventory
- Chapter 43Operating Language-Model Applications
- Chapter 44Evaluation Suites for LLM Applications
- Chapter 45Tracing and Observability for LLM Applications
- Chapter 46Guardrails in Production
- Chapter 47Data Validation in Practice
- Chapter 48A Minimal Pipeline in Code
- Chapter 49Monitoring in Code
- Chapter 50Data Contracts
- Chapter 51Online Experiments for Models
- Chapter 52Explainability in Production
- Chapter 53Fairness Monitoring
- Chapter 54Human-in-the-Loop Operations
- Chapter 55Sustainability
- Chapter 56Organisational Practices
- Chapter 57Costs of Language-Model Applications
- Chapter 58Choosing Tools
- Chapter 59Case Study: The Silent Column Change
- Chapter 60Case Study: Training–Serving Skew
- Chapter 61Case Study: The Audit Question
- Chapter 62Case Study: A Language-Model Assistant
- Chapter 63Building an MLOps Roadmap
- Chapter 64A Checklist
- AppendixExercises
- GlossaryGlossary
Who this book is for
ML engineers, data scientists and platform engineers responsible for models in production.
Details
| Format | PDF + web edition |
|---|---|
| Length | 84 pages (approx.) |
| Chapters | 64 |
| Level | Intermediate to Advanced |
| Language | English |
| Topic | AI Engineering |
| Author | Latent Library Editorial |
Keep reading