search

methlabUZH / automagic

Automagic
GNU General Public License v3.0
94 stars 32 forks source link

ICA Weights look strange #32

Closed behinger closed 3 years ago

behinger commented 4 years ago

This is what the ICA winv looks like if I plot the EEGset I receive after automagic: grafik

These are the one I get, when I overwrite the EEG.icaact (winv etc.) with the EEG.etc.beforeICremove.icaact etc.: grafik

Note that in the latter component 1 & 5 were marked for deletion. But besides that, something went wrong somewhere. The first set of components does not look at all like a usable ICA-decomposition.

I am using the normal automagic cloned from github yesterday

ksgfan commented 3 years ago

Hi The problem arose when recomputing 'icawinv' using eeg_checkset after removing bad ICs. icawinv = pinv(EEG.icaweights * EEG.icasphere) and icawinv = pinv(EEG.icaweights(good_comps,:) * EEG.icasphere) are not the same.

We fixed it - now we use 'icawinv' computed directly in pop_subcom - and updated the master branch. Best, Dawid

behinger commented 3 years ago

ok cool :-)