ENH: add spatio-temporal clustering example for sensor data #1642
Conversation
mne/channels.py
Outdated
dengemann
force-pushed
the
spatio_temporal_sensors
branch
2 times, most recently
from
74345d6 to
60367ea
Compare
|
@agramfort ready for merge from my end. |
|
just curious, how are these neighbor templates generated? is it possible that there is a general way that could be used independent of system? |
@t3on take a look at the related fieldtrip documentation: http://fieldtrip.fcdonders.nl/template/neighbours they are manually edited. |
|
ah, i see. they do a combo of automated and manual correction. the manual part is used for addressing the edges of the neighborhoods. if it were fully, it would be ideal, no trailing template files. cool, it looks good. |
|
@t3on yes, exactly. I think they have spent some efforts to get it right. IIRC initially they pursued an automated approach which turned out insuccient. It woulld be cool to have a magic bullet here, but if it would be that failsafe no one would use these templates. |
|
I'm a little bit surprised a Delaunay triangulation won't work -- you could even use Dijkstra distances afterward to make it based on spatial distance. I assume they tried this sort of thing, though. And if other people have already developed a sufficient solution (manual fixing), I'm fine with that. |
There was a problem hiding this comment.
@agramfort we had decided aleady earlier to only allow for dropping channels if preloaded. But there was a bug. We only checked for Raw istanced, not for Epochs. I corrected this, because it made the example fail. I then updated the test, becaus they failed as a consequence of actually putting the droppring API with the preload constraint into practice. Ok?
There was a problem hiding this comment.
@agramfort we had decided aleady earlier to only allow for dropping channels if preloaded. But there was a bug. We only checked for Raw istanced, not for Epochs. I corrected this, because it made the example fail. I then updated the test, becaus they failed as a consequence of actually putting the droppring API with the preload constraint into practice. Ok?
ok
dengemann
force-pushed
the
spatio_temporal_sensors
branch
from
03ad320 to
67754a5
Compare
minor fixes update tests to deal with preload when dropping channels cleanup cleanup2 cleanup viz
fix setup.py edit manifest fix attempt setup.py more setup + py3k fixes cleanup, add Fieldtrip credits, update reference and whats new address suggestions from hangout remove montages.py
dengemann
force-pushed
the
spatio_temporal_sensors
branch
from
67754a5 to
5309c53
Compare
|
@agramfort Travis is happy. I addressed your suggestions. I'd like to merge this before #1619. It will be less painful to rebase this way. |
ENH: add spatio-temporal clustering example for sensor data
|
Thanks @agramfort for your final edits! |
5 participants
MNE Python has been capable of this for at least three months. Time to expose it :)
It finds 6 clusters using an F-test, which is justified here as we do not have any ties on a first level analysis
(permuiting condition order does not affect hypotheses).
Here's two of them:
For conveniences, I've added the neigho(u)r files from fieldtrip, slightly renamed to make a difference.
Thanks for the attendees of our Paris workshop who motivated me to give this example, especially @matthew-tucker
cc @agramfort