Feat/interpolate channels #480
Conversation
fix event list in tests
moabb/datasets/preprocessing.py
Outdated
| # Trick: mark these channels as bad | ||
| raw.info["bads"].extend(missing_channels) | ||
| # ...and use mne bad channel interpolation to generate the value of the missing channels | ||
| raw.interpolate_bads(origin=(0, 0, 0.04)) |
There was a problem hiding this comment.
Hi @gcattan , I really like your idea about channel interpolation to match electrodes of two datasets.
Is there a reason why in interpolate_bads you have selected origin (0, 0, 0.04) ? Option "auto" will fit a sphere to headshape based on the existing electrodes and determine its radius and origin.
There was a problem hiding this comment.
Hi! I am not sure it will work if the montage is disabled. But let me try.
There was a problem hiding this comment.
@Sara04 I added a try-catch, so we first try with the 'auto' parameter, and default to (0, 0, 0.04) in case of ValueError. This happens, for example, when there is no montage information. The braindecode pipeline fails, but I don't think the issue originates from this PR.
There was a problem hiding this comment.
Thanks @gcattan ! Yes, it doesn't seem that it fails due to this PR.
There was a problem hiding this comment.
All checks passed. @Sara04 FYI
There was a problem hiding this comment.
Sure. Is it ok to merge this PR then?
There was a problem hiding this comment.
For me it is OK. @sylvchev , @bruAristimunha do you have some other comments?
gcattan
force-pushed
the
feat/interpolate_channels
branch
from
9337792 to
0f8b864
Compare
…n/moabb into feat/interpolate_channels
…n/moabb into feat/interpolate_channels
Head branch was pushed to by a user without write access
|
@bruAristimunha @Sara04 |
|
Done :) |
This PR modifies the
match_allmethod inside base paradigm, which set-up the paradigm to be compatible with a list of datasets.It is now possible to select the intersection of the channels in all datasets, or their union.
For example, if you pass two datasets:
d1 -> Cz, Fz
d2 -> Cz, C4,
then the
channelsattribute inside the paradigm can contain either: Cz or {Cz, Fz, C4}.In the case you selected the union of all channels, then missing channels in dataset d1 or d2 will be interpolated using mne.
This interpolation is done directly in the
RawToEpochstransformer.@bruAristimunha FYI