Probablistic Programming and Bayesian Methods for Hackers | Hacker News Reader

Probablistic Programming and Bayesian Methods for Hackers

camdavidsonpilon.github.io

281 pointscnivolle27 comments

I worked my way through the book a month ago. It's very practical and understandable. The ipython notebook format makes it extremely easy to play with the code without worrying about any setup at all. And having it on github made it extremely easy to fix and clarify things as I went.

Did you find a use for it at your job? One of the challenges is to make something that is new and useful to other people.

I think whether it's applicable depends on your job. Two immediate ideas for me were modeling query cost (I work on a high performance server at work, and one of its responsibilities is to guarantee certain rates of queries to certain users) and anomaly detection (for monitoring that server). In my workplace we already have better tools for some of these things, but it's nice to know I could whip up my own if I spent a little time to think about the random variables.

I'm glad you enjoyed it =)

And please, I'm curious about a response to Don_Draper's question: have you been able to apply either Bayesian inference, or Bayesian thinking?

How long did it take you to work through everything?

FYI for those in Boston this weekend: http://pydata.org/bos2013/

Cam (the author) will be presenting an overview: http://pydata.org/bos2013/abstracts/#46

Strange, got a malware warning :

Danger: Malware Ahead! Google Chrome has blocked access to this page on camdavidsonpilon.github.io. Content from ctan.yazd.ac.ir, a known malware distributor, has been inserted into this web page. Visiting this page now is very likely to infect your Mac with malware.

Hmmm, is this still happening to you? Is it happening to anyone else?

Not to me. Perhaps a sign of a dodgy Chrome extension inserting content into the DOM.

Another interesting platform for working with probabilistic programming is Bher/Church:

http://projects.csail.mit.edu/church/wiki/Church

and Oleg's and Ken Shan's fantastic OCaml library Hansei:

http://okmij.org/ftp/kakuritu/index.html

Pedantic nit: probablistic vs probabilistic....

>Pedantic nit: probablistic vs probabilistic....

Your nit reminded me of this: http://web.archive.org/web/20080516195038/http://cs-people.b...

"Lower Bounds for Probably-istic Polynomial Time" an extremely funny read :)

Are there any plans at all for a dead-tree version, for us old timers? I love what little I've read of this book, but reading on screens just doesn't work for me.

Yes, there are plans. There is currently a PDF version, floating somewhere in the Github repo. This version is out-of-date, and is only a draft. A more physical book may be coming mid-fall though.

I admire the approach and allowing anyone to contribute, but for me to delve into things, I personally prefer something I can take offline and browse and take some notes on...PDF really is the best bet and would be helpful for me in this case...

I have only had success with the stock Android browser for viewing ipython notebooks on a mobile platform. However, for some reason the "A" in P(A) in Chapter 1 does not display. Does anyone know why this would be happening?

This book also serves as excellent guide to ipython/matplotlib visualisation so I would recommend it for this even if you are not interested in primary subject matter.Very well done on all counts.

Not that I have any complaints (the opposite is true), but why is the cover art an illustration of the Toronto skyline?

A few reasons:

1. I lived in Toronto when I started writing, plus the graphic designer currently lives there.

2. It's nice to see a Canadian connection ;)

3. The results of Bayesian inference sorta look like a skyline, see http://i.imgur.com/bF4YrBQ.png

Cool, thanks!

I've lived in Toronto for the past seven years, and I'm always happy to see it mentioned in technology and entrepreneurship contexts.

Does the author have any blog post or plan to write one about this part: "After some recent success of Bayesian methods in machine-learning competitions"? It will then be easier to translate Bayesian Inference to real world problem for noobs like me. I was trained in mathematical Bayesian Inference in an Econometrics class but I never did anything practical with it.

Hi, Author here. I don't have any plans to write that blog post, but I'll can provide some resources:

1. http://timsalimans.com/observing-dark-worlds/ 2. http://timsalimans.com/winning-the-dont-overfit-competition/

You mention: "trained in mathematical Bayesian Inference in an Econometrics class but I never did anything practical with it" . This book tries to bridge that, from theory to practice. The examples I provide tend to be very practical.

Many thanks for those links.

Here's another (same comp, MCMC models took top 2) http://blog.kaggle.com/2012/12/19/a-bayesian-approach-to-obs...

Reply on news.ycombinator.com