Looks good except for the C coding style which I think should conform more closely with the numpy conventions. Thanks for adding this, it should be very helpful to other folks down the road.

I should also note that I believe Mark will add support for masks at the loop level at some point, so that is worth keeping an eye on.

Chuck's review looks pretty thorough, not much to add! In the master branch, there is support for masks at the loop level already. It works transparently with the existing inner loops like in these examples, but one day it will be worth adding some loops specialized for performance like I described here:

http://projects.scipy.org/numpy/ticket/1901

You can see the exposed details here:

https://github.com/numpy/numpy/blob/master/numpy/core/include/numpy/ufuncobject.h#L17
https://github.com/numpy/numpy/blob/master/numpy/core/include/numpy/ufuncobject.h#L145

Thanks charris! I only just read PEP 7 last weekend; hadn't realized I was violated python c code conventions.

Are there any numpy C code conventions beyond those in PEP 7? I looked online at the numpy docs and googled a bit, but I couldn't find anything.

I've pushed changes I knew how to make and responded inline.

The C conventions aren't really written down at this point, they mostly consist of me complaining about new code. Because I reformatted a good bit of the code back when, they are sort of de facto conventions ;). I should probably write something up.

I think I made it Py3K compliant, based on the examples charris mentioned. But I'm not sure at all. Help/insight would be appreciated.

Getting close, it looks much better.

I'll check this with python 3 if you don't want to do that.

Thanks for all the checking!

I just pushed a new version. I think I got everything you mentioned.

If you could do the Python 3 check that'd expedite things; I'm traveling for the next week, and won't have the time to setup a python 3 virtual env on my computer, load everything up, and check it until I get back.

With the modifications, the module compiles and runs under python 3 (with a deprecation warning).
Impressions:

• npspam is an unfortunate choice of name
• a small test/demo program would be useful
• you should indicate where the resulting module ends up so newbies don't lose it.

Danke! I changed the errors you caught, and thank you very much for checking.

I also added the comment in the setup.py file for how to build the library in place. In general, there are comments in the setup.py file for how to use setp.py and where the results will be located.

There are small demos inside each section. Hopefully I'm not being too dense, but I'm not really sure why adding something would be useful. They can just cut, paste, and build the libraries as they are to play around with it themselves. The code is specifically structured so that can be done with a minimum of fuss.

And I agree that npspam is not the best name. I can't think of anything I like a lot better, though, that makes any sense in the context of the tutorial. pretty_unicorn and walrus_guano would be awesome names, yet likely inappropriate for a logit function. Statisticians don't have that much humor. Perhaps for a package containing modified chebyschev polynomials or rayleigh-ritz method algorithms. I'm open to suggestions for an alternative name to npspam.

I'd just use something dull and uninspired, like npufunc.

Dull and uninspired works for me. For public libraries, at least.

Pushed.