Refactored project as installable package

[khornlund]

Re: #6

Summary of changes:

• Added setup.py file to enable setuptools install
• Moved partition_spline.npz to resources submodule, and modified code to use pkg_resources to access it.
• Moved resources loaded by tests to tests/resources/
• Added a utility method _get_device to adaptive_test.py to handle selection of torch device (I was getting errors trying to run the tests without the number on the end of 'cuda:{num}').

Notes:

• I couldn't figure out the nosetest way to skip the CUDA tests if no GPU is available. Is this possible?
• Perhaps we should include some more detailed examples in the README?

Cheers, Karl

[jonbarron]

Very nice work! Thanks for doing this. I also couldn't figure out how to skip CUDA tests, and just got used to the idea of them failing when there's no GPU available, which isn't perfect but doesn't seem too bad.

I agree that more detailed examples are needed. I can make a "hello world" regression demo script, though I've never written a pytorch optimization loop so it may be a bit weird. That should resolve some of the confusion in the other thread about how to include parameters.