ENH: Add minimum-norm based model for virtual sensors

mne-tools / mne-python

MNE: Magnetoencephalography (MEG) and Electroencephalography (EEG) in Python

https://mne.tools

BSD 3-Clause "New" or "Revised" License

2.7k stars 1.31k forks source link

ENH: Add minimum-norm based model for virtual sensors #6085

Closed larsoner closed 1 year ago

larsoner commented 5 years ago

The map_data functions should become public somehow for mapping data between runs.

I think we can do it pretty easily actually by incorporating it into maxwell_filter (you get many things for free this way) but Alex does not like it yet. For discussion at the sprint.

Todo:

[ ] Allow maxwell_filter to use a minimum norm approach
[ ] Make _map_meg_channels public
[ ] Allow _map_meg_channels to use multipole moments instead of minimum norm
[ ] Speed up minimum norm approach (if possible)

jasmainak commented 5 years ago

maybe something like: https://gist.github.com/jasmainak/756cf02dce82ab2dc5c2c69722dd13d1

larsoner commented 5 years ago

Also as suggested by @jjnurminen _map_meg_channels could also use multipole moments instead of minimum norm. I think ideally one should be able to use either transformation approach (minimum norm or multipole moments) in both maxwell_filter and in whatever ends up being the public version of _map_meg_channels.

See also https://github.com/mne-tools/mne-python/pull/4491 where @kingjr wanted things to be faster -- the multipole moment method could end up being faster, but either way we can tackle speed as part of this issue.

larsoner commented 1 year ago

I think we have this sort of thing with https://github.com/mne-tools/mne-python/pull/7366 and epochs/evoked.as_type

FWIW the transformation is linear so if you want it to be really fast, do it once with an EvokedArray(np.eye(info["nchan"]), info) or so and then do the matmul to data as needed