机器学习设计模式

　　《机器学习设计模式（影印版）》中的设计模式针对机器学习中反复出现的问题给出实践和解决方案。作者为来自谷歌的三位工程师，他们整理了已证实的方法，帮助数据科学家解决整个机器学习过程中的常见问题。这些设计模式将数百位专家的经验转化成直接、易懂的建议。在这本书中，你会找到关于数据和问题表示、操作化、可重复性、可复现性、灵活性、可解释性和公平性的30种模式的详细解释。每个模式包括对问题的描述、各种可能的解决方案，以及针对你的情况选择技术的建议。你将学习：
　　·识别和减轻在训练、评估以及部署机器学习模型时的常见挑战
　　·为不同类型的机器学习模型表示数据，包括嵌入、特征交叉等
　　·针对具体问题选择合适的模型类型
　　·使用检查点、分发策略和超参数优化，建立一个鲁棒的训练循环
　　·部署可扩展的机器学习系统，通过它你可以再训练和更新以反映新的数据
　　·为用户解释模型的预测结果，确保模型公平地对待用户
　　·提高模型的准确性、可复现性和弹性

　　Valliappa（Lak）Lakshmanan是谷歌云数据分析和人工智能解决方案的全球负责人。
　　
　　Sara Robinson是谷歌云团队的开发者和倡导者，专注于机器学习。
　　
　　Michael Munn是谷歌的机器学习解决方案工程师，他帮助客户设计、实现和部署机器学习模型。

　　★“具有优秀且多样的示例，这本书是任何数据科学家或机器学习工程师理解复杂机器学习问题的经过验证的解决方案的必读书籍。”
　　——David Kanter MLCommons执行董事
　　
　　★“如果你想在构建机器学习解决方案的过程中少一些疤痕，肿块／瘀伤，Lak、Sara和Michael会支持你。”
　　——WiIIGrannis 谷歌云CTO办公室董事总经理

Preface
1．The Need for Machine Learning Design Patterns
What Are Design Patterns?
How to Use This Book
Machine Learning Terminology
Models and Frameworks
Data and Feature Engineering
The Machine Learning Process
Data and Model Tooling
Roles
Common Chauenges in Machine Learning
Data Quality
Reproducibility
Data Drift
Scale
Multiple Objectives
Summary

2．Data Representation Design Patterns
Simple Data Representations
Numerical Inputs
Categorical Inputs
Design Pattern 1： Hashed Feature
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 2： Embeddings
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 3： Feature Cross
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 4： Multimodallnput
Problem
Solution
Trade-Offs and Alternatives
Summary

3．Problem Representation Design Patterns
Design Pattern 5： Reframing
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 6： Multilabel
Problem
Solution
Trade-Offs and Alternatives
Design Pattern 7： Ensembles
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 8： Cascade
Problem
Solution
Trade-Offs and Alternatives
Design Pattern 9： Neutral Class
Problem
Solution
Why It Works
Trade-Offs and Alternatives
Design Pattern 10： Re alanang
Problem
……
4．ModeI Training Patterns．．．
5．Design Patterns for Resilient Serving
6．Reproduability Design Patterns
7．Responsible AI
8．Connected Patterns
Index

　　Whols This Book For?
　　Introductory machine learning books usually focus on the what and how of machine learning （ML）. They then explain the mathematical aspects of new methods from AI research labs and teach how to use AI frameworks to implement these methods. This book， on the other hand， brings together hard-earned experience around the "why"that underlies the tips and tricks that experienced ML practitioners employ when applying machine learning to real-world problems.
　　We assume that you have prior knowledge of machine learning and data processing.This is not a fundamental textbook on machine learning. Instead， this book is for you if you are a data scientist， data engineer， or ML engineer who is looking for a second book on practical machine learning. If you already know the basics， this book will introduce you to a catalog ofideas， some of which you (an ML practitioner） may rec-ognize， anc：give those ideas a name so that you can confldently reach for them.If you are a computer saence student headed for a job in industry， this book will round out your knowledge and prepare you for the professional world. It will help you learn how to build high-quality ML systems.
　　What's Notin the Book
　　This is a book that is primarily for ML engineers in the enterprise， not ML scientists in academia or industry research labs.
　　We purposefully do not discuss areas of active research-you will fmd very little here，for example， on machine learning model architecture (bidirectional encoders， or the attention mechanism， or short-circuit layers， for example） because we assume that you will be using a pre-built model architecture (such as ResNet-50 or GRUCell）， not writing your own image classification or recurrent neural network.
　　Here are some concrete examples of areas that we intentionally stay away from because we believe that these topics are more appropriate for college courses and ML researchers：
　　ML algorithmsWe do not cover the differences between random forests and neural networks，for example. This is covered in introductory machine learning textbooks.Building blocksWe do not cover different types of gradient descent optimizers or activation functions. We recommend using Adam and ReLU-in our experience， thepotential for improvements in performance by making different choices in these sorts of things tends to be minor.
　　ML model architectures
　　Ifyou are doing image classiFication， we recommend that you use an off-the-shelf model like ResNet or whatever the latest hotness is at the time you are reading this. Leave the design of new image classification or text classification models to researchers who specialize in this problem.
　　Model layers
　　You won't fmd convolutional neural networks or recurrent neural networks in this book. They are doubly disqualified-first， for being a building block and sec-ond， for being something you can use off-the-shelf.
　　Custom training loops
　　Just calling model. fit(）in Keras will Fit the needs of practitioners.In this book， we have tried to include only common patterns of the kind that machine learning engineers in enterprises will employ in their day-to-day work.As an analogy， consider data structures. While a college course on data structures will delve into the implementations of different data structures， and a researcher on data structures will have to learn how to formally represent their mathematical properties，the practitioner can be more pragmatic. An enterprise software developer simply needs to know how to work effectively with arrays， linked lists， maps， sets， and trees.It is for a pragmatic practitioner in machine learning that this book is written.