Adding a function to distributions.py

[wesm]

And testing out the git workflow.

There's a global git setting that trims trailing whitespace in files, tends to be good practice in general-- I have emacs do it automatically for me, but as you can see it creates diff cruft. Up to you...

[apatil]

Thanks for the function, Wes. Are you also planning on implementing inverse_wishart_cov_like and inverse_wishart_cov_expval?

We do actually have a script to remove trailing whitespace, just aren't as vigilant about using it as we should be.

Cheers,
Anand

[wesm]

Sure, I can do that for consistency's sake. I think you can set up a pre-commit hook to do it automatically on commits, there are various recipes floating around out there:

http://snipplr.com/view/28523/git-precommit-hook-to-fix-trailing-whitespace/

separately there's a git config flag that warn you if there's trailing whitespace.

I'm still learning the pymc codebase but I will periodically make contributions over the next few years probably, so bear with me :)

[apatil]

Great, thanks! I think we'll wait for the full set of functions before pulling. Thanks also for the precommit hook, will have a look. Good luck and let us know on the ML if anything is confusing.

Anand

[wesm]

Got it, I'll do that soon and get back to you.

Do you guys have a general testing policy for the distribution functions?

[apatil]

Yes, but I didn't want to stretch the friendship. See

http://github.com/pymc-devs/pymc/blob/master/pymc/tests/test_distributions.py#L968

and also

http://github.com/pymc-devs/pymc/blob/master/pymc/distributions.py#L65

needs to be modified to fully integrate it.

[wesm]

No problem at all. As a quid pro quo you might catch me reformatting code to fit in inside the 80 character margin, hope you don't mind terribly =)

[wesm]

Anand-- on reviewing the code again, I realized that I had things backwards-- what I was looking for was an rinverse_wishart function parameterized using the precision matrix instead of the scale/covariance matrix. I got confused because the function signature is rinverse_wishart(n, Tau), and Tau means precision elsewhere (like in rwishart).

I can tweak the naming convention and docs to make it clear what the distinction is.

So there are some issues:

• the commit I made a pull request with earlier today is incorrect, I'll fix it in my branch. If you're OK having that revision in there then there's no problem
• do you have a preference for what to call this other form of the distribution?
• I noticed that rwishart / rwishart_cov give different answers for the same random seed-- I wrote a unit test to compare the output, I changed rwishart_cov to call rwishart for the time being (this will all be in a separate commit), not sure how you'd like to address this