Mpl2 display

[bmcfee]

This PR updates the test suite for matplotlib 2.0 compatibility, and fixes #233 .

Note: the display still works on 1.5, but for testing purposes, we now use 2.0 (and 2.0 default styles).

I had to change a couple of small things in the test setup to get things working -- notably, removing tests/__init__.py (tests does not need to be a module), and added an extras_require for testing.

[bmcfee]

Note: this PR needs to get merged before any of the others, because display tests are currently breaking everything else.

[bmcfee]

It looks like the source separation metrics are getting some minor errors now. I can't tell if it's due to numerical precision or an actual change in the data. Note: i didn't change any data here, only locations.

@dawenl got a minute to investigate?

[bmcfee]

follow-up: I don't see any of these separation errors when i run the tests locally. Only notable difference I see between my env and travis is that I have openblas-numpy, and travis is vanilla. Spoke too soon; it does indeed fail locally as well.

I suspect that issue is orthogonal to this PR though.

EDIT: confirmed that these separation failures are independent of the changes in this PR. I recommend we make a separate PR to fix the separation fixtures.

[craffel]

EDIT: confirmed that these separation failures are independent of the changes in this PR. I recommend we make a separate PR to fix the separation fixtures.

@faroit or @ecmjohnson may be the best people to ask since they touched the codebase most recently.

[faroit]

@craffel I will have look this weekend

[bmcfee]

@faroit @rabitt pointed out that this might be due to randomness in the bss_eval functions.

If so, this can be resolved by seeding the RNG before each test function executes. librosa does this by a helper function here, example here.

[bmcfee]

As mentioned over in #236, I think this one's good to go. The only remaining failures are on separation metrics.

A quick CR would be nice, since this PR does change a little of how the tests are executed (beyond just the display submodule). It does not change any functionality though.

[craffel]

So the -w tests tells nosetests it's running within the tests folder, and this is why you removed the tests prefix from all the paths?

[bmcfee]

Yep. This way you can also execute from within the tests directory, and the tests don't have to know that they live within the tests/ directory of the repo.

[craffel]

Cool.

[craffel]

Looked it over, looks good. Just want to reiterate how awesome the display module is!!

[bmcfee]

@craffel thanks! merge away?

[bmcfee]

@faroit did you have a chance to look at the source sep fixtures?