Automatically run 2to3 when installing on Python 3

[dan-blanchard]

I added a couple lines to setup.py to make 2to3 get run automatically by Distribute on Python 3 because we were running into some issues installing this on 3.3. I just followed the recommended steps here.

[benhamner]

Looking at the Travis build, this prevented pandas from being installed -

https://travis-ci.org/benhamner/Metrics/builds/5992585

ERROR: test.test_auc (unittest.loader.ModuleImportFailure)

ImportError: Failed to import test module: test.test_auc
Traceback (most recent call last):
File "/usr/lib/python2.7/unittest/loader.py", line 252, in _find_tests
module = self._get_module_from_name(name)
File "/usr/lib/python2.7/unittest/loader.py", line 230, in _get_module_from_name
import(name)
File "/home/travis/build/benhamner/Metrics/Python/ml_metrics/test/test_auc.py", line 4, in
import ml_metrics as metrics
File "ml_metrics/init.py", line 1, in
from quadratic_weighted_kappa import *
File "ml_metrics/quadratic_weighted_kappa.py", line 4, in
import pandas as pd
ImportError: No module named pandas

Can you fix it?

[dan-blanchard]

I think this may be a misconfiguration of Travis, rather than an issue of my changes breaking something, because I noticed that the build details say that it can't find requirements.txt, which is what tells it to install pandas in the first place.

It might be necessary to change

requirements = [x.strip() for x in open("requirements.txt")]

to

requirements = [x.strip() for x in open(os.path.join(os.path.dirname(__file__), "requirements.txt"))]

so that it looks for requirements.txt in the right place, but this shouldn't have changed from what I did.

I'll make the change on my end and see if your automatic travis build passes then.

[dan-blanchard]

Well, that doesn't appear to have cut it. I think it's something you have to change with the Travis config, because of the error message "Could not locate requirements.txt. Override the install: key in your .travis.yml to install dependencies."

[dan-blanchard]

Alright, after a lot of messing around with it, I believe I've fixed the YAML file. Annoyingly, there doesn't seem to be a simple way to get numpy working with Python 3.3 on Travis, so I gave up on getting this testing working for 3.3. It just tests 2.7, but the tests all pass now.

[dan-blanchard]

I hate to be a pest, but this PR should be good to go now, and I'd really like to not have to install this using a custom git repository URL, so could you please pull this?

[benhamner]

Thanks for the reminder, merged!

[dan-blanchard]

Could you please update the version of ml_metrics that is on pypi to include these fixes? You may have to bump the version number up to get it to recognize the changes. Currently when I try to install via pip with python 3 it does not actually run 2to3 automatically like it's supposed to.