Various ICA warnings and failures

Sameer Walawalkar sameer at ANDREW.CMU.EDU
Fri Feb 23 20:39:53 CET 2007


thanks, This fixed the rank problem.

With the new pop_runica, runica worked very fast on the ICA

but the ICAs don't seem to make sense at all. As I said in the earlier
email, they all look the same except for overall scale issues.

best,
sameer




On Fri, 23 Feb 2007, arno wrote:

> The rank problem might be related to this issue on 64-bit machines
>
> http://sccn.ucsd.edu/pipermail/eeglablist/2007/001675.html
>
> Best,
>
> Arno
>
> Sameer Walawalkar wrote:
>> Hi,
>>
>> This mail has 3  questions related to implementing ICA is fieldtrip.
>>
>> 1)
>> I am now getting the following warnings from runica
>> I am getting these messages while doing ICA on a datafile which contains
>> sss MaxFiltered data (Elekta Neuromag).  When I visually inspect the
>> independent components, they seem fairly identical to me. Their basic time
>> structure seems to be same, just the scale is different.
>>
>> Prior to this, I had run ICA on the same datafile but with no MaxFiltering
>> (or SSP projection for that matter).
>>
>> Final training data range: -0.00443438 to 0.00355995
>> Computing the sphering matrix...
>> Warning: Matrix is close to singular or badly scaled.
>>          Results may be inaccurate. RCOND = 2.156998e-29.
>>> In runica at 774
>>   In componentanalysis at 331
>> Starting weights are the identity matrix ...
>> Sphering the data ...
>> Beginning ICA training ...
>> Data has rank 286. Cannot compute 307 components.
>> Warning: Matrix is close to singular or badly scaled.
>>          Results may be inaccurate. RCOND = 2.155705e-29.
>>> In componentanalysis at 437
>> total time in componentanalysis 160.6 seconds
>>
>> 2)
>> binica on this datafile just hangs, after one learning, it sends nan and
>> just hangs.
>>
>>
>>
>> 3)
>> Fast ICA is giving me following errors
>>>> compRR_One_fastica_sss = componentanalysis(cfg,dataRR_One);
>> Warning: cannot determine whether the FASTICA toolbox is present
>>> In fieldtrip/private/hastoolbox at 121
>>   In componentanalysis at 176
>> Warning: adding FASTICA toolbox to your Matlab path
>>> In fieldtrip/private/hastoolbox at 145
>>   In componentanalysis at 176
>> Warning: Name is nonexistent or not a directory: fastica.
>>> In path at 113
>>   In addpath at 100
>>   In fieldtrip/private/hastoolbox at 146
>>   In componentanalysis at 176
>> selecting 307 channels
>> baseline correcting data
>> concatenating data..............................
>> concatenated data matrix size 307x120030
>> starting decomposition using fastica
>> Number of signals: 307
>> Number of samples: 120030
>> Calculating covariance...
>> ??? Error using ==> componentanalysis
>> call to fastica failed
>>
>> But when I do
>>>> which fastica
>> /mnt/condor2/sameer/KarmaCond/DataNeuroMagCondor2/MegSoftwareCondor2/FastICA_2.5/FastICA_25/fastica.m
>>
>>>> exist('fastica','file') (as seen in hastoolbox)
>>
>> ans =
>>
>>      2
>>
>>
>> thanks,
>> sameer
>>
>> p.s: I am currently running varimax on the sssed data and binica on the non
>> sssed to see what happens.
>>
>
>



More information about the fieldtrip mailing list