Elements Of Statistical Learning: now free pdf | Hacker News Reader

Elements Of Statistical Learning: now free pdf

www-stat.stanford.edu

183 pointswglb28 comments

May I also recommend a related book, also a free PDF: http://www.inference.phy.cam.ac.uk/mackay/itila/ "Information theory, inference, and learning algorithms" by David Mackay. This book is not a pure machine learning book but it is a very fun read.

http://www.inference.phy.cam.ac.uk/mackay/itila/Potter.html

I haven't read it, but this comparison has convinced me of its worth.

Anyone want to compare and contrast the two? Is one better overall? Do the books have different strengths and weaknesses?

The McKay textbook is more about Bayesian inference and a few neural network chapters.

Hastie is more of a statistical approach with statistical rigor.

I'm mainly interested in Machine Learning algorithms. I've read the first 75 pages of this book, and skimmed most of Mackay, and I prefer this one. It goes into a lot more detail on the performance of different algorithms, how the handle sparse data, error rates and high dimensionality.

McKay's looks like a good book, but appears more applicable to pure information theory rather than ML specifically.

A lot of the Mackay book is on information/coding theory and while it will deepen an existing understanding of ML, its probably a roundabout introduction. ESL is a much better intro, especially for someone looking to apply ML.

That said, it is meant for people who are comfortable with math/stats; its much more statistics oriented than, say, Mitchell's book. But they do a good job of explaining things in non-math language. This book does a good job of exposing high level ML concepts (e.g. bias-variance tradeoff) but still teaches a lot of the standard methods & tools.

These are actually two of my favorite books on the subject and I can't recommend them both enough.

I noticed the R packages for various functions in the text can be found here http://www-stat.stanford.edu/~tibs/ElemStatLearn//Rfun.html .

Their names are pretty catchy; I'll admit I want to use 'gradient boosting' and 'Lasso and elastic-net regularized generalized linear models'.

We just covered Neural Nets in my machine learning class, and the professor made some comments about how far a good name can take you. Neural Nets are better than other learning algorithms in some ways and worse than others, but the name gives you the sense that you are learning the secrets of how the brain works. That may or may not be true, but it sure has served as good branding for the Neural Networks algorithm.

I have read this book and it is awesome.

Alright then, added to my reading list.

Agreed. It's great both as a reference and as a teaching/learning tool.

I had read a couple of pages in the book on linear regression and was very impressed by the way concepts had been explained. Although approach is very mathematical, I will recommend this book to anybody serious about Machine Learning

The section on linear regression is extra awesome.

several years ago, i took a class from prof tibshirani using this book and was absolutely blown away. if you don't have a deep stats background (as i didn't), you might find it scarily dense, but wow there's some good stuff in there.

I also used this book for a data mining class that I took about two years ago. The book is very thorough but the math can be dense. We didn't actually end up using it in class very much, but it was a good reference book for the topics it contains.

I am starting to learn statistics. But I have zero knowledge in the subject. I have picked up a copy of Headfirst Statistics. Is that a good one? Any recommendations?

For a good and concise introduction to basic statistics I'd recommend: "Principles of Statistics" by M.G. Bulmer

http://www.amazon.com/Principles-Statistics-M-G-Bulmer/dp/04...

It's kind of old, but it helped me a lot...

Edit: It's also on Google Books, if you want to take a look

"All of Statistics" by Larry Wasserman is very good. It is traditional, fairly mathematical statistics, but intended for a general audience. (It is particularly popular with computer science people.)

It is also extremely concise, perhaps too much so occasionally for pedagogical ease. It really does come pretty close to covering "all of statistics" in 442 pages, an impressive feat. But as you can imagine it certainly doesn't beat around the bush much. So if you're looking for something with multiple applications, covering the same material from different perspectives, etc., or just want more hand holding, this may not be for you.

I love that the icon for technically difficult sections is Munch's The Scream.

The visualizations are very nice.

Looks like a very useful text.

Since this book is so math heavy, can anyone recommend a free pdf book that can be used as a math intro for this? I like the book, but I get a little over my head with the math in some sections.

Don't know about math intro book, but I can't think of "free pdf" and machine learning without also thinking about

http://www.autonlab.org/tutorials/

Probability for Data Miners, for example, has much of the basic math you need for machine learning algorithms, for example.

Another useful resource might be the 20 part lecture series on Machine Learning by Professor Andrew Ng for CS 229 in the Stanford Computer Science department. Youtube playlist URL - http://www.youtube.com/view_play_list?p=A89DCFA6ADACE599