Don't freeze version of packages

[alxkolm]

Please, don't freeze versions of packages

'numpy==1.12.1',
'scikit-learn==0.18.1',
'scipy==0.19.0'

[nicodv]

After reading https://blog.miguelgrinberg.com/post/the-package-dependency-blues, I've moved to ranges of dependency package versions:

install_requires=[
        'numpy>=1.13.0, <1.14.0',
        'scikit-learn>=0.18.0, <0.19.0',
        'scipy>=0.19.0, <0.20.0',
    ]

[shapiromatron]

scikit-learn has a few more recent versions, could this range be increased?

[nicodv]

@shapiromatron Unfortunately, scikit-learn 0.20 introduces some breaking changes to k-modes. I'm holding off for a bit, but feel free to try it and file a PR. :)

[shapiromatron]

Thanks @nicodv; I just wanted to make sure it really was required. I'll take a look at see what I can do....

[trevorstephens]

What breaks Nico? gplearn hasn't seen any issues since the RNG changed in I think 0.18 and I believe we mostly use the same sklearn utils

[nicodv]

I seem to recall the sklearn compatibility tests have to be rewritten, @trevorstephens : https://github.com/nicodv/kmodes/blob/master/kmodes/tests/test_common.py

I've always had to customize that file entirely, because kprototypes takes an extra argument (categoricals) to the fit/predict methods and that breaks sklearn compatibility. I'd still like to comply as much as I can, though.

[trevorstephens]

Ah yes that makes sense. I saw something vaguely similar from Py-earth in the mailing list saying some of his tests failed due to extra args as well:

I'm working on updating py-earth for some recent changes in scikit-learn and cython. It seems like check_estimator has been significantly improved, and I'm working through making py-earth compliant with it. I've hit the following issue, though. It seems check_estimator tests score_samples using only X as an argument, and py-earth's score_samples requires y as well. So, my question is: must score_samples work with just X (and therefore maybe I should just remove it from py-earth) or is it okay to have a score_samples that requires y, and I should try to find a workaround for check_estimator?

You might want to pile onto this PR with some suggestions for your package as was suggested for py-earth: scikit-learn/scikit-learn#8022