Book Review: Learning From Data

Learning From Data is the companion text for an introductory machine learning course on edX with the same name, taught by one of the authors, Yaser Abu-Mostafa. Unlike Data Science for Business and Applied Predictive Modeling, Learning from Data focuses on machine learning theory and is targeted to an audience with a strong math background (especially in linear algebra.). As such, each page of Learning from Data tends to take a long time to read so despite the books subtitle, "a short course," it isn't an especially quick read.

Coming in at less than 200 pages, Learning from Data isn't a comprehensive text that gives the theory and mathematical details of every machine learning algorithm around. It has only 5 chapters and much of the content deals with the theory of learning and considerations when applying machine learning rather than algorithms and techniques themselves. Chapter 1 covers machine learning basics, the perception as an example of a basic classification algorithm and discusses the theoretical feasibility of learning with computation. Chapter 2 expands on the first by discussing generalization, VC dimension and the bias-variance trade-off. Chapter 3 is the only one that specifically covers algorithms. It focuses on linear models in detail, including linear regression, logistic regression, gradient descent and using non-linear transformations with linear models. Chapter 4 discusses over-fitting, regularization and validation, while chapter 5 wraps up with three general considerations to keep in mind when applying machine learning: Occam's Razor, sampling bias and data snooping.

Learning from Data is a great book for those looking to take a dive into machine learning theory, but the material can be hard to follow. Much of the content could be boiled down to a few sentences that capture the practical implications of the theory; readers interested in arming themselves with an array of practical tools should look elsewhere. Purchase of the book does give you access to e-Chapters that cover other algorithms in detail, including support vector machines and neural networks. The authors intend to update the e-Chapters from time-to-time to keep up to date with the latest machine learning trends, which may be a valuable resource going forward.

I give Learning from Data 4 out of 5 stars. It is a very good book for what it is but, but not a book I would read for fun.