[FieldTrip] numcomponent problem in ica

KatrinH Heimann katrinheimann at gmail.com
Thu Sep 25 15:53:14 CEST 2014

Hey experts,

I have a problem with an ICA. What I am basically doing is running an ICA
on a big dataset and then removing the identified components from smaller
datasets of the same subject. Due to interpolated channels I do reduce the
components using numcomponent. But there seems to be a problem with this:

My code is

cfg = [];

cfg.channel = {'all'};

cfg.numcomponent = 126

comp = ft_componentanalysis(cfg,obs_data_clean1);

save (strcat(sb,'comp_all') , 'comp')

and matlab writes:
the input is component data with 126 components and 129 original channels
detected   0 visual artifacts
the input is component data with 126 components and 129 original channels
processing trials

just as expected!

then I want to use this ica for cleaning a smaller dataset so I wrote:

cfg = [];

cfg.numcomponent = 126;

cfg.unmixing  = comp.unmixing;

cfg.topolabel = comp.topolabel;

comp_1     = ft_componentanalysis(cfg, obs90_data);

 the program again tells me:

the input is raw data with 129 channels and 80 trials

selecting 129 channels

baseline correcting data

scaling data with 1 over 126.842103

not concatenating data

starting decomposition using predetermined unmixing matrix

the call to "ft_componentanalysis" took 2 seconds and required the
additional allocation of an estimated 574 MB

then I remove the component I identified before as blinking

cfg = [];

    cfg.component = [1];

    obs90_data_ica_cleaned = ft_rejectcomponent(cfg, comp_1, obs90_data);

 save (strcat(sb,'obs90_ica_cleaned') , 'obs90_data_ica_cleaned')

and suddenly matlab says:

baseline correcting data

removing 1 components

keeping 128 components

processing trials

processing trial 80 from 80

Why does is speak of 129 (128+1) components??? My comp and comp_1 objects
only have 126 components!!! Is it doing the right thing?

Thanks again for your help!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20140925/db01131e/attachment-0001.html>

More information about the fieldtrip mailing list