I just received my copy of "Collective Intelligence" in the mail this morning, and immediately sat down and starting reading. I've been reading for the past several hours, non-stop. It's that good. Perhaps the best part is the chapter at the end of the book that neatly and concisely sums up all the major ML algorithms: neural networks, SVM, kNN, k-means, decision trees, Bayesian classifiers, etc. etc. This chapter alone is worth the price of the book. I only wish it were longer (and it's not a short book).

What would you say is the level of the book? I am intrigued by your enthusiasm, but I think I'd be wasting my time if I went reading an introductory book.

Introductory. I don't have a particularly formal background in discrete math or ML, but I was able to absorb most of the content.

I'd say it's introductory, up to the chapter it discusses SVMs. Than it becomes incomprehensible, unless you have prior knowledge of SVMs and kernel methods.

You deserve a commission, Eric. I just bought "Collective Intelligence" based on your review.

Ditto. Here's an Amazon link, for the lazy: http://www.amazon.com/Programming-Collective-Intelligence-Bu...

As mentioned in this thread http://news.ycombinator.com/item?id=3564873, there's another new book "Machine Learning in Action" from Manning uses python instead of R: http://www.manning.com/pharrington/

If you're using Python, what are you using for your stats libraries?

R is, generally, the go to language for stats work I've found. It's certainly used a lot in the financial world for statistical modelling, and all the libraries in it are well tested which is a big plus in my book.

Scikits.statsmodels [1] is the main statistical and econometric library for Python. It is usually used with pandas [2], which provides nice interfaces to your data, particularly for time-series. You can also call R functions with the rpy2 bridge [3]; pandas provides a higher-level API for using rpy2, though last I checked it's not fully fleshed out yet.

Wes McKinney, the developer of pandas, writes a blog [4] that provides a good look at data analysis in Python, particularly if you're in finance.

1. http://statsmodels.sourceforge.net/

2. http://pandas.sourceforge.net/

3. http://rpy.sourceforge.net/rpy2.html

4. http://blog.wesmckinney.com/

Check out enthought http://www.enthought.com for "everything and the kitchen sink". Numpy and scipy will generally have you covered, and often wrap low-level libraries written in e.g. FORTRAN and LAPACK. YMMV but I've used python exclusively in scientific computing as well as "web startup" data analysis environments. I love being able to write my data analysis and web server code in the same language. :)

I'm also partial to R, but Python does have stats support via NumPy and (especially) SciPy: http://docs.scipy.org/doc/scipy/reference/stats.html

When I was at the Broad, at least, there was heavy use of both R and Python among the statistically savvy.

Interesting - I'm not well versed on the Python stats ecosystem. It's a side effect of working in finance where every place I've worked already has large well tested libraries in C/C++/C#.

Here's a directory of stats/ML-related Python libraries: http://aliquote.org/memos/2011/02/07/python-for-statistical-...

Raw level: numpy

More high-level statistical functionality: scipy.stats (built on top of numpy)

Even more high-level: scikit.statsmodels (built on top of scipy)

I liked "Programming Collective Intelligence," as it fairly clearly introduces how the algorithms operate by implementing them from scratch.

I found the coding style a bit non-pythonic, though (two space indentation?), and in practice you'd be better served using one of the many ML libraries (e.g., scikit-learn [1]), which aren't introduced.

[1] http://scikit-learn.org/stable/index.html

The two space indentation is probably to fit more code on the page without softwraps.