[Eeglablist] Problems with ICA decomposition

Velu Prabhakar Kumaravel velu.kumaravel at unitn.it
Sat May 22 14:02:19 PDT 2021


Dear list,

I encounter the following issue in a couple of noisy datasets with ICA
analysis. For some reason, the number of ICA activations/weights is lower
than the total number of channels. This situation throws an error running
the adjusted-ADJUST method (Leach et al., psychophysiology 2020).

*Some insights on the data:*

                data: [*117*×250×197 single]
              icaact: [*116*×250×197 single]
             icawinv: [117×116 double]
           icasphere: [117×117 double]
          icaweights: [116×117 double]
         icachansind: [1×117 double]
            chanlocs: [1×117 struct]

There are 117 channels in the data with 116 IC activations. Perhaps, the
data is not full-rank?

While I'm waiting for the developers to provide a solution for this, I
wonder if there's any generic workaround for this scenario.

If it helps, here's the error message (obviously related to index mismatch):


*"Index in position 1 exceeds array bounds (must not exceed 116).*





*Error in MARA_extract_time_freq_features (line 174)
plot(nobumps,1:50,resids_pxx_matrix(good_ics(gg),:))Error in
adjusted_ADJUST (line 347)cwb=MARA_extract_time_freq_features(EEG); "*


Thanks in advance,

Sincerely,

Velu Prabhakar Kumaravel, Ph.D. Student
Center for Mind/Brain Sciences,
University of Trento, Italy


More information about the eeglablist mailing list