[WIP] add conda env file

[agramfort]

it would be nice to test this file in our CIs

then we would just recommend to do:

conda env create -f environment.yml

to install a working MNE environment.

EDIT: DO NOT MERGE, need to figure out how to fix traitsui so we don't need to skip those tests.

[larsoner]

also needs nilearn

I can push and make Travis use this sometime today hopefully

Actually we should make ci-helpers support this

[larsoner]

See astropy/ci-helpers#232

[codecov-io]

Codecov Report

Merging #4523 into master will decrease coverage by 69.25%.
The diff coverage is 88.88%.

@@             Coverage Diff             @@
##           master    #4523       +/-   ##
===========================================
- Coverage   83.61%   14.35%   -69.26%     
===========================================
  Files         349      349               
  Lines       65439    65463       +24     
  Branches    10155    10152        -3     
===========================================
- Hits        54715     9397    -45318     
- Misses       7849    55822    +47973     
+ Partials     2875      244     -2631

[larsoner]

... and now astropy/ci-helpers#233

[larsoner]

@agramfort the current version doesn't work because of conflicts.

Even doing:

conda create -n test2 python=3.6
source activate test2
conda install -c menpo menpo numpy scipy matplotlib mayavi

I get the conflict. Can you get one to work?

I ask because once we have a working env we can probably do conda env export to get the right yml file, or at least something close.

If you have Py3k + Mayavi in an env, can you try the command to see what the yml file looks like, maybe paste it somewhere?

[ktavabi]

Question about this approach. Is the recommendation to install mne-python stable or development version in the virtualenv?
If it helps at all I have a 2.7 env file that works without any of the backend or Mayavi excitement. But I see you guys are transitioning to 3.6.

[larsoner]

Stable is the recommended version for users

Hold off on sharing the file for now, I want to see if we can get 3.6 to work

[ktavabi]

Thanks, @larsoner FUQ do you know if it's possible to use different dev branches across different virtual envs?

[larsoner]

Not that I know of when using a single directory, but you could always git clone to two different directories like and python setup.py develop from each of them as desired

[agramfort]

I just pushed a working env for osx

can someone try it on linux?

[larsoner]

Yes I can try today. astropy/ci-helpers#233 is merged so I can then try to get a Travis build running with it.

[agramfort]

yes !

[larsoner]

Other than the hilarious error ModuleNotFoundError: No module named 'mne' the env works. I'll try making Travis use it

[agramfort]

great thanks yeah I did not include it for travis :)

[larsoner]

yeah I did not include it for travis :)

The python setup.py develop should take precedence I think. We'll find out...

[agramfort]

we don't need nitime nose-timer nose-faulthandler anymore?

[larsoner]

we don't use nose anymore

[larsoner]

had to rebase to master to resolve conflicts

[larsoner]

The GUI stuff doesn't work because of a Py3k bug in traitsui, but when I update it to use pip it doesn't fix it, and when I update to git master, it breaks in a different way. So we're a bit stuck.

In other words, pytest mne/gui/ fails, so @agramfort you have a broken env :(

[larsoner]

(but other than that, this should be working at this point)

[larsoner]

Pushed a commit that should skip those tests (which is not good) to at least see if we can find green

[larsoner]

Keeping track of sklearn versions and API changes is a bit of a nightmare

[larsoner]

@agramfort the environment2.yml seems to have a problem on Windows -- it installs sklearn 0.16, which means most new stuff in decoding can't be used :(

It also just crashes (without any output!) on Travis. I'm a bit stuck.

[larsoner]

Apparently environment3.yml has the same problem on Windows

[larsoner]

... but when I use conda env create -f environment3.yml on my Windows VM it uses sklearn 0.19. I don't get it.

[larsoner]

Scratch that, it's just the 2.7 env that has sklearn 0.16, the 3.6 env has 0.19

[larsoner]

There is enthought/mayavi#528, so maybe we have a chance at PyQt5 (which was holding back matplotlib)

[agramfort]

@larsoner can you split the pytest migration and the datetime fix to other PRs to help us review.

then it will "just" be changes in CIs scripts.

thanks a lot

[larsoner]

can you split the pytest migration and the datetime fix to other PRs to help us review.

Sure, hopefully this weekend but maybe not until Monday

[larsoner]

@agramfort I'll start a new PR once the other changes are merged.