Elements of Statistical Learning | Hacker News Reader

Elements of Statistical Learning

www-stat.stanford.edu

79 pointssonabinu15 comments

I highly endorse this book. Best explanation of linear regression I've ever read. I think a good litmus test for stats books is whether they explain how minimizing squared error corresponds to maximizing probability given Gaussian-distributed errors, and, of course, whether they acknowledge the eternal light of Bayes shining in the background of all things.

I've gone through this book. I found it useful for theory, but not useful for gaining an appreciation of how the models work in practice. Most practitioners will appreciate a more applied book than this one.

Agreed. Any suggestions for those applied books? I haven't found any yet.

I found Machine Learning for Hackers to be good as a practical, case study based book for learning to apply and understand what is going on in several machine learning problems.

http://shop.oreilly.com/product/0636920018483.do http://www.amazon.com/Machine-Learning-for-Hackers-ebook/dp/...

It's definitely a practical approach. There's a lot of explanation but, despite coming from two PhD candidates, it certainly did not read like an academic paper.

It's also case study based, not necessarily algorithm based.

As a bonus, I used that book to teach myself R. I don't think it's meant to be an intro to R tutorial, but it worked for me.

All three are positives in my book.

A good practical book is Machine Learning in Action" http://www.amazon.com/dp/1617290181/

Yep, Pattern Classification by Duda, Hart and Stork:

http://www.amazon.com/Pattern-Classification-2nd-Richard-Dud...

It is very pragmatic, including algorithms for many machine learning and artificial intelligence topics (from fitting functions for classification or regression purposes to search processes). The authors have a strong industrial background (in addition to the academic).

ESL is pretty much the go-to text for statisticians interested in machine learning. The math in it might be a bit more difficult than some of the pages you see on HN, but you won't get much better than this for a survey-course on machine learning from a statistician's perspective.

interesting. i've been going through gilbert strangs book and courses. this looks like a nice accompaniment.

I found Christopher Bishop's book to have a good deal of knowledge in it. Plus, he compliments many ideas with graphical figures, which is a big plus. http://www.amazon.com/Pattern-Recognition-Learning-Informati...

Also, if you needed more information about optimization methods all of Stephen Boyd's books are really good, just check out his entire website for information. http://www.stanford.edu/~boyd/

other free drafts/full open content books

http://www.ics.uci.edu/~welling/teaching/273ASpring10/IntroM...

http://alex.smola.org/drafts/thebook.pdf

http://www.p-value.info/2012/11/free-datascience-books.html

http://web4.cs.ucl.ac.uk/staff/D.Barber/textbook/090113.pdf (webserver's on verge of falling over, maybe we can stagger requests)

and i think some others are: https://news.ycombinator.com/item?id=4672930

The one from UCI is a very rough draft, shdn't have included that

great book, but you need to have a very high level math background to understand most of it (in my opinion)

Looks like completely normal math to me. It's not like they're trying to shove category theory down your throat or anything.

A useful book! Look just like what I needed. If there's something heavier about it, it's the terse encyclopedic feeling of it, it's not a gentle introduction in this sense.

Well that is not very helpful. If the last math class you took was 20 years ago, ESL probably seems rather intimidating. The good thing is that most of the book requires only calculus and linear algebra. By brushing up on those topics first, you could be in a position to understand most of the basics of machine learning.

"If the last math class you took was 20 years ago, ESL probably seems rather intimidating."

Actually, only ten years ago. :-) But I do need a refresher. Just started working on it recently. I feel actually smarter today than I felt back then, so I don't think will hurt that much.

"The good thing is that most of the book requires only calculus and linear algebra. By brushing up on those topics first, you could be in a position to understand most of the basics of machine learning."

Yeah, I've noticed. The scope of the whole book is vastly overblown for my needs, my application (NLP-related) won't probably need more then the basic stuff, but any good material at my disposal helps.

Reply on news.ycombinator.com