Probabilistic Programming and Bayesian Methods for Hackers
nbviewer.ipython.orgGah, this reminds me so strongly of undergraduate Economics, where all the textual information and explanation was completely clear to me and all the manipulation of formulae was completely impenetrable to me. I deeply regret not studying maths to a higher level. There is an entire language of symbols and notation that I don't understand, and it's frustrating because I imagine the symbols and notation are trivial in difficulty to learn compared to the underlying logic of the operations.
I feel the same way whenever I think about quantum mechanics, fluid dynamics, thermodynamics, medical terminology, probability/statistics, functional programming, APL, and internal combustion engine repair manuals. For me, the biggest barrier to grokking something is almost always terminology.
I'm not sure if not knowing is better or worse than forgetting. I used to really excel at multivariable calculus and now I have to ask people on physics forums or wolfram alpha to solve basic integrations. I really, really wish they made Cliffs Notes for people who already learned something. For example I think every college course should have on the order of a one hour review video that covers everything. Maybe a hybrid of spaced interval learning and Khan Academy.
> If frequentist and Bayesian inference were programming functions, with inputs being statistical problems, then the two would be different in what they return to the user. The frequentist inference function would return a number, representing an estimate (typically a summary statistic like the sample average etc.), whereas the Bayesian function would return probabilities.
Wouldn't a "frequentist inference function" typically return a probability distribution, not a summary statistic?
I just love that IPython-driven style of explanation becoming popular. I wish there will still be appearing more of this. Maybe something a little more advanced: I guess Bayesian methods is something most of us already know.
I hate being one of the, "Hey this isn't the original title, downvote!" but the submitted title here doesn't do justice to the already relevant title of the OP, and the breadth of OP's material:
"Probabilistic Programming and Bayesian Methods for Hackers"
This is truly a labor of love, and at version 0.1, very readable. I got through about half of it on the subway ride to work.
Every time I see "Bayesian Probability," I mentally re-load http://plover.net/~bonds/cultofbayes.html
Love the statistics, hate the cult.
A really great start. Keep it up!
Would be helpful to have a version that doesn't depend on internet connectivity... e.g. so I can read it on my iPad offline.
Does iPython easily allow saving as a pdf?
"printing" it as a pdf works sort of ok. Also, you should be able to view it in an ipython notebook running locally offline on a traditional computer.
Also want to mention that there are 7 completed chapters on the github repo. Seems like an amazing resource.
https://github.com/CamDavidsonPilon/Probabilistic-Programmin...
The movie "12 angry men" sounds like a Bayesian inference, starts with some assumption and then changes over new information.