Re: [Biosig-general] findclassifier [was Re: Biosig-general Digest, Vol 2, Issue 1]

[Singh, H. <Har...@wa...>]

Dear Alois

Thanks for the quick help. The error message is the same, even after removing ng parameter. 


Just to check things I tried another option to get the classifier using train_sc, even then the error doesn't change.

 CC = train_sc(f,HDR.Classlabel(1:10));
??? Error using ==> train_sc at 88
length of data and classlabel does not fit

Now the size of f when I use the first 10 trials is 30000*128 and as soon as I parse the parameters to the train_sc, line 86 says  

sz = size(D); 
if sz(1)~=length(classlabel),
        error('length of data and classlabel does not fit');
end; 

%% D is f so sz(1) is 30000 which can not be equal to length(classlabel) = 10

It will be great if you can spare some moments to please help

Thanks
Warm regards
Harsimrat


-----Original Message-----
From: Alois Schlögl [mailto:alo...@tu...] 
Sent: 02 July 2008 11:22
To: Singh, Harsimrat
Cc: bio...@li...
Subject: re: findclassifier [was Re: [Biosig-general] Biosig-general Digest, Vol 2, Issue 1]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

The error message does not indicate a memory problem. The problem is the use of the ng parameter which defines the cross-validation procedure. If all elements in ng are the same, either the training or the test set is empty, and you do not get any result.

I suggest to omit the ng parameter in
CC   = findclassifier(f, HDR.TRIG(1:10), HDR.Classlabel(1:10), seg,
flag, 'LD3');
This is the same as ng=[1:length(HDR.TRIG)]' and is equivalent to a leave-one-trial-out X-V procedure.

Alois




Singh, Harsimrat wrote:
> Dear BioSig Users
>  
> I have been trying to use BIOSIG methods for the analysis of Dataset I 
> of the BCI Competition data. The challenge is data is too large and 
> comes in one file BCI2005_I.gdf which contains both the training (278) 
> and the test trials. I use sload to load the data. because of large 
> memory it requires, i want to use first 10 trials to do the classifier
>  
> % each trial is 3 sec long with sampling freq of 1000 Hz f = 
> bandpower(s(1:30000,:),HDR.SampleRate,[10,12;16,24],1);
> f(f<-10)=NaN;
>  
>  
> seg  = reshape(1:8*HDR.SampleRate,HDR.SampleRate/5,8*5)'; % define segments, typical 1/10s .. 1/5s
> flag = seg(:,1)>(3*HDR.SampleRate);     % flag of possible segments
> ng   = ceil([1:length(HDR.TRIG(1:10))]'/length(HDR.TRIG(1:10))*K); % K_fold XV; K=10;
> CC   = findclassifier(f, HDR.TRIG(1:10), [HDR.Classlabel(1:10),ng], seg, flag, 'LD3');
>  
> it gives me an error that length of the data and classlabel does not fit.
> ??? Error using ==> bci4eval at 102
> BCI4EVAL: size of data and size of Classlabels does not fit
>  
> Second option is to try and remove the last 368 trials using sigviewer and save the file as gdf, but sigviewer is not able to save the file also. 
>  
> Can anyone help me with a way around this problem.
>  
> Thanks in advance
> Harsimrat
>  
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFIa1bIzSlbmAlvEIgRAhrMAJ9Hwzwtq0omhLbRH60l84yXVRgBIACfUjhh
LjVNC0Dd4et/80HYlSME744=
=YZnv
-----END PGP SIGNATURE-----