[.ca] The Elements of Statistical Learning, Second Edition: ... (ISBN 0387952845)

Pedagogical Disaster:
The Hastie book was used at our major university to teach data mining and statistical learning. The students in this graduate-level course included people with Masters and PhD degrees, as well as post-docs. Most people work in the field of bioinformatics, so have a pretty good grasp of complex topics and computer science, as well as mathematical algorithms. The overall rating from the course was a D-, which is one of the worst ratings for a book that was used on campus (out of hundreds). The text was hard to follow, confusing in many sections, and tough to teach from. It does cover a lot of ground, which is a benefit. But apparently the ability to do justice to clearly cover such breadth is a challenge that 20 really smart people couldn't figure out. Maybe individuals with a strong background and understanding in one or more of the areas covered by the book can do well by this item, but from a teaching/learning perspective there is at least one group of folks out here who would have done better with some other alternative.

One of the Essential Books on Modern Machine Learning:
This book is a miracle of clarity and comprehensiveness. It presents a unified approach to state of the art machine learning techniques from a statistical perspective. The layout is logical and the level of math is appropriate for applications-oriented engineers and computer scientists, as well as theorists. Sections where the book does need to go into heavier mathematics are clearly marked and generally optional. I found the book very easy to read, but at the same time very comprehensive. The book provides a very illuminating counterpoint to other books that promote the Computational Learning Theory (COLT / kernels / large margins) viewpoint of modern machine learning. Many of the same techniques such as boosting and support vector machines are discussed, but are motivated in different ways. Appropriate regularization is seen as the key to avoiding overfitting with complex models, rather than margin maximization. Mathematically you often end up solving the same problem, but personally I usually find the statistical approach much more direct and intuitive. This book is a nice follow on to introductory pattern recognition texts such as Duda and Hart, though it can be read without any prior pattern recognition knowledge. It provides a nice mix of theory and paractical algorithms, illustrated with numerous examples. An essential element of your machine learning library!

Excellent introduction to statistical learning:
This book is an excellent survey of the huge area of statistics / computer science called statistical learning. The discussion is interesting and accurate, but not too theoretical. It is the best book to date for a general audience with a reasonable math/stat background. One of the strengths is the wide variety of topics covered; it is very comprehensive. If there is a weakness, it is that depth is limited. Plenty of references are provided for further study, and the authors maintain a website. Recommended as a reference or a starting point for an applied statistician or mathematician, or as a text for a first course in the subject.

Covers many topics breifly:
I was already familiar with many of the topics covered in this book, but had to do a double take when reading about familiar concepts. Unfortunately, the authors' unique perspective is not presented in a way that is benificial to the reader. I would strongly suggest another book as a reference or introduction to this material.

Counter to review from Sep 8:
The review from September 8 expresses an opinion which is the exact opposite of mine, and is worded so strongly that I have to object. I gave a course using the book to bioinformaticians, most of them with a computer science background, and found the book exceptionally well prepared and suitable for a graduate course. The book serves the dual purpose of an introduction and a reference. An especially nice feature is how the authors explain the relationships and differences between different methods. By doing so, they provide context which I have not seen in any other book on this subject. The book is a very nice combination of basic theory and performance evaluation on data from a wide variety of domains and it is quite up-to-date. It has a well developed website going with it and the graphical material can be obtained electronically from the publisher. The book is an outstanding contribution to the field.