MRG, DOC: note on using picard for same solution as other algos

[sappelhoff]

Picard can be used to obtain the same solution as with FastICA, Infomax, or extended Infomax --> and apparently (according to Picard 😉 ) converges faster.

This sounds like a win for me, so I added this to the docs. WDYT?

@larsoner @cbrnr @agramfort @mmagnuski @pierreablin

Info taken from README: https://github.com/pierreablin/picard

Previously discussed in gitter, Thanks @cbrnr for making me aware of it.

[cbrnr]

Looks good - I think this is something that people are not very much aware of. We should advertise this on the EEGLAB mailing list 😄.

[larsoner]

https://21307-1301584-gh.circle-artifacts.com/0/dev/generated/mne.preprocessing.ICA.html#mne.preprocessing.ICA

[sappelhoff]

however, digging into this, I also saw that picard is currently not tested for Python 3 --> mind-inria/picard#22

[agramfort]

Really happy to see other people saying the same thing :)

we'll catch up on testing and release / doc with @pierreablin ASAP

[cbrnr]

CIs show some package conflicts (traits) - do we fix these before merging?

[GuillaumeFavelier]

I would suggest to update with master (0b3ad03) because I don't see any -n used in .travis anymore.

[cbrnr]

@sappelhoff can you rebase?

[cbrnr]

Alright, now seeing the errors described in #8040. Any ideas?

[GuillaumeFavelier]

I compared with a recently merged PR and pyface has been bumped from 6.1.2 to 7.0.1. That could explain the new requirement?

5 days ago: https://travis-ci.org/github/mne-tools/mne-python/jobs/709093476#L909
this PR: https://travis-ci.org/github/mne-tools/mne-python/jobs/710681059#L906

[cbrnr]

Probably. But which package is causing this runtime error then? There is no conda error when installing packages.

[cbrnr]

Testing in #8042.

[GuillaumeFavelier]

I think it's because pyface 7.0.1 is available on conda only recently?

Details

[cbrnr]

You mean they might have overlooked the dependency on traits? conda info pyface shows traits as a dependency without any version restriction (but maybe it should be >=6).

[GuillaumeFavelier]

It's just a guess since it's the first big change I notice. But this could explain why there is no error during installation.

[cbrnr]

Yes. And according to https://docs.enthought.com/pyface/changelog.html#release-7-0-1, pyface >= 7 requires traits >= 6 (which is not available in conda yet).

[cbrnr]

traits 6.1.0 is already in the channel (https://anaconda.org/anaconda/traits/files), so maybe this issue will resolve itself.

[cbrnr]

Some package apparently pins traits to <6 because conda update traits says that the latest version is already installed (5.2.0).

[cbrnr]

Maybe someone who actually knows what they are doing could help, because I don't use conda and I can't figure out what the conflict is (@hoechenberger maybe?).

[cbrnr]

Also, let's fix this issue in #8040 or #8042 instead of cluttering this PR.

[hoechenberger]

Great move, @sappelhoff! @agramfort was waiting for somebody to step in and finally file this kind of PR ;) Good to see this happen!! We've been using Picard in my old lab and it worked really well. It's also our default now in the mne-study-template :)

[agramfort]

thx @sappelhoff