MRG: Minor fixes to testing framework

[larsoner]

This PR:

1. Gets rid of some nose cruft
2. Moves to standard pytest slow-test markers
3. Fixes our decorators for sklearn that had incorrect versions
4. Makes AppVeyor a little bit nicer

Closes #4518.

[larsoner]

@agramfort CIs look like they're headed toward happiness

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@cf9ff8b). Click here to learn what that means.
The diff coverage is 100%.

@@            Coverage Diff            @@
##             master    #4541   +/-   ##
=========================================
  Coverage          ?   87.02%           
=========================================
  Files             ?      349           
  Lines             ?    65609           
  Branches          ?    10174           
=========================================
  Hits              ?    57099           
  Misses            ?     5630           
  Partials          ?     2880

[agramfort]

remove nose?

[agramfort]

besides my concern of keeping some nose in travis LGTM

thx heaps @larsoner

@massich wanna have a look?

[larsoner]

No we still do a lot of nose assert imports. It will take a separate PR with thousands of changed lines to fix that

[agramfort]

how about having our assert_* functions in fixes.py so we don't need to change every single line for now?

[larsoner]

Only hundreds of lines then :) Probably better for a follow up PR though

[agramfort]

yes ok.

[massich]

LGTM

[massich]

how about having our assert_* functions in fixes.py so we don't need to change every single line for now?

I would do that as intermediate step if we really need to remove the nose dependency. But as end goal I would change all assert_* to pytest like asserts.

Either way I would also do them on a different PR. So +1 in merging this one as it is.
BTW, @larsoner any idea why macos travis is not happy?

[agramfort]

@larsoner you need to rebase

[larsoner]

@agramfort done

[agramfort]

is the Makefile and the make test target already switched to pytest ?

if so +1 for MRG

[larsoner]

Yep, @massich did that in a previous PR and I tweaked the values here based on the attribute/mark switch

Just pushed a commit to fix a rebase error, will merge once the CIs come back happy

Then we can figure out #4523 :) Maybe the first thing to do is just put the existing Travis 2.7 config in environment.yml. That way we'll have something that works on CIs, even if it doesn't work well for users. Then we can iterate on just modifying that file (and/or adding a separate 3.5/6 one) without needing to mess very much with Travis or AppVeyor.

[larsoner]

Going to have to rebase again. @agramfort okay to merge once CIs are happy?

[agramfort]

yes ok to merge when Cis are green