[Eeglablist] runica

[arno]

For those interested in this discussion, there is a follow up in a 
bugzilla bug report.

https://sccn.ucsd.edu/eeglab/bugzilla/show_bug.cgi?id=317

Arno

Sven Hoffmann wrote:
> Hi all,
>
> i have a question concerning the new eeglab version.
> i have run runica (and also pop_runica) several times, and if i remove 
> one or several components, or rereference to average reference, and then 
> recompute runica, the same number of components is computed as before. 
> This appeared a bit strange to me since the removal of components or 
> average reference reduces the rank of the data. Having a look at the 
> revision history i found that the data is converted to double precision 
> prior  infomax calculation. So i tried the rank() matlab function of 
> matlab and it seems that applying it to single or double precision leads 
> to different results. Below is the difference in the pop_runica code:
>
> pop_runica (v5.03)                                                   
> 482         tmprank = rank(tmpdata(:,1:min(3000, 
> size(tmpdata,2))));                     
>
> pop_runica(v6.0b)
> 555        tmprank = rank(double(tmpdata(:,1:min(3000, 
> size(tmpdata,2)))));   
>
> May be this is a bug? I would suggest to convert the data to double 
> precision following rank() calculation.
>
> Cheers, Sven.
>