Making package pip installable

[mjstevens777]

This new version of the package should be installable via pip. I tested with python 3 on OS X and a fresh install of ubuntu.

• I reworked the setup.py script to require numpy and cython.
• I added a package name so that it installs under pygco instead of UNKNOWN
• I put the gco_src files (without MATLAB) into the repo permanently so that the separate Makefile step is not necessary.

[amueller]

The gco license prohibits redistribution, see the GCO_README.TXT that you included.

[mjstevens777]

Thanks for the fast review. I saw the section on citations, but I missed the part about distribution, thanks for catching that. I updated the process to automatically pull the sources instead of baking them in. I admit, my approach is pretty hacky so feel free to reject it.

[amueller]

I mean my original approach was pretty hacky and of questionable legality ;)

[amueller]

That actually doesn't look to bad. If you want brownie points, the right thing to do would be to have this tested on travis - and this repo should have travis anyhow.

This thing is very low on my list of priorities, tbh, so if you don't feel like working on CI, I'm happy to merge without it.

Have you looked at OpenGM? They have a very comprehensive suite of wrappers and implementations.

[amueller]

my change was insubstantial, but I needed to enable travis on the repo for it to run, and then needed to make a commit so it actually runs.

[amueller]

sweet, looks good! merge?

[mjstevens777]

Sounds good. Thanks for your help on this! Go ahead and merge

OpenGM and pystruct are great, I just have trouble installing them on OS X. They also don't seem to support label costs. I am planning on raising a 10-line change for that as well, would you like that here or in a separate PR?

[amueller]

separate PR please. And I'm not actively maintaining pystruct right now, which is a bit of a shame :-/