Pytest

[RomainBrault]

#30

[coveralls]

Coverage remained the same at 100.0% when pulling 5420785 on RomainBrault:pytest into ac1f099 on scikit-learn-contrib:master.

[rth]

Thanks for this PR!

• Please also replace as needed other mentions of nosetests in particular in the readme and in setup.cfg.

• You can also replace nose setup by pytest in appveyor.yml

[rth]

It might still be worth keeping nose in the dependencies (in addition to pytest), as e.g. older versions of numpy.testing will use it, if I remember correctly.

[RomainBrault]

I kept nose in install.sh and appveyor.yml and .gitignore; and added pytest to install.sh and appveyor.yml. No other mentions of nose should be present.

[RomainBrault]

Also, I'm wondering why requirements.txt includes nose? I think neither pytest or nose should be in.

[rth]

Thanks for the update; a few more comments are below.

It might still be worth keeping nose in the dependencies (in addition to pytest), as e.g. older versions of numpy.testing will use it, if I remember correctly.

I kept nose in install.sh and appveyor.yml and .gitignore; and added pytest to install.sh and appveyor.yml. No other mentions of nose should be present.

I'm not sure what should be done there; maybe @lesteve would know...

[rth]

That's not necessary, it's specific to your editor (vim I imagine) and should be set in your global .gitignore

[rth]

--ignore doesn't work with --pyargs unfortunately pytest-dev/pytest#3287

[RomainBrault]

I was a bit confused. Actually pyargs is mandatory to get the path right easily and ignore is superfluous

[lesteve]

It might still be worth keeping nose in the dependencies (in addition to pytest), as e.g. older versions of numpy.testing will use it, if I remember correctly.

Very unfortunately scikit-learn 0.19.1 has a hard dependency on nose (try importing sklearn.utils.testing without nose installed) so it may be the case that we still need nose (as a dependency not to run the tests) in the CIs for scikit-learn-contrib.

numpy.testing is irrelevant, basically nose was still used in a weird edge case (maybe numpy had something like numpy.testing.assert_equal that was using nose). The thing is that there is no actual need to use numpy.testing.assert_equal (or whatever the actual function was) and you can emulate it like we do in scikit-learn with unittest.

[NicolasHug]

Is there anything keeping from merging this?
It could make the badges go green again :)

[rth]

Is there anything keeping from merging this?

Finding someone who has merge rights probably..

cc @glemaitre

[NicolasHug]

@amueller ?

[glemaitre]

Merging this one. I think that we should have more update regarding this template.

[rth]

Thanks @glemaitre ! A few thoughts on what could be simplified in #40 ..

[glemaitre]

I am working on it. I will make a todo list there.