[Eeglablist] Complex number in EEG.icaweights due to rank defficiency?

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Thu Feb 25 10:25:46 PST 2021


Dear list members,

I ran into a strange error while investigating one of the user's data.
EEG.icaweights is complex. How does it happen? Is this also due to the rank
deficiency? Is this another failure mode in running ICA when input data is
rank deficient? After some investigation, I found that it is most likely so.

There are 32 ch and 32 ICs. However, the electrode data seems practically
rank-deficient.

>> eigs(double(EEG.data)*double(EEG.data)', [], 32)

ans =

   1.0e+07 *

    6.8695
    3.7526
    2.3744
    2.0239
    1.3012
    0.7264
    0.5364
    0.4846
    0.4016
    0.3233
    0.2683
    0.2477
    0.2139
    0.1958
    0.1738
    0.1549
    0.1355
    0.1140
    0.0860
    0.0768
    0.0187
    0.0017
    0.0013
    0.0011
    0.0009
    0.0008
    0.0008
    0.0008
    0.0008
    0.0008
    0.0007
   -0.0000

The last eigenvalue is -4.4977e-10. Reference status says 'average', so the
initial reference electrode was not included in re-referencing, hence
practically rank deficient by one.

When I use 'pca', 31, the problem was solved, so most likely this is due to
rank deficiency.

Makoto


More information about the eeglablist mailing list