Hi Darren and Shelly
I had this problem previously, it's not serious (thanks to Karl for an
off-list answer).
I found an apparently singular covariance matrix C at the deconvolution step
which causes an error in spm_PEB from line 351 in spm_peb_ppi, i.e.
C=spm_PEB(Y,P).
This turned out to be due to a sparse matrix inversion setting in spm_PEB
and is not a serious problem but can be resolved by setting exp(-32) to
exp(-16) in
line 371 (of spm_PEB) ==> C = inv(C + speye(length(C))*exp(-32));
This makes sure that C is not too close to singular and inversion proceeds
without warnings.
Hope this helps
Alexa
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of d gitelman
Sent: 07 August 2007 00:47
To: [log in to unmask]
Subject: Re: [SPM] PPI Analysis: SPM5 Produces Matlab Warning
Hi Shelly:
Could you please provide a bit more information
1) How many conditions for each session and which ones are you modeling.
Also how many trials per condition per session?
2) Does the warning occur for every PPI calculation. If you use a
multisession spm i assume you are calculating 2 PPI's per subject.
3) what is the version of spm.m, of spm_peb_ppi.m and of spm_regions.m. You
can used the attached script to return the version strings. Put the script
somewhere in your matlab path. Start matlab. cd to the spm5 directory and
type retVer('spm.m'), etc.
darren
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping)
> [mailto:[log in to unmask]] On Behalf Of Shelly Fitzgerald
> Sent: Friday, August 03, 2007 6:03 PM
> To: [log in to unmask]
> Subject: [SPM] PPI Analysis: SPM5 Produces Matlab Warning
>
> Dear SPM'ers,
>
> I have a question regarding PPI using SPM5. Thus far I've
> conducted the following steps:
>
> 1. run 1st level stats (2 sessions per subject; each session
> modeled separately in the same design) 2. extract VOIs for
> each session 3. calculate PPI regressors for each VOI
>
> While calculating the PPI regressors, however, I receive the
> following matlab warning:
>
> Matrix is close to singular or badly scaled. Results may be
> inaccurate. RCOND=1.266417e-20.
>
> I checked the PPI and VOI data and there are no NaN values as
> others have reported. Does anyone know what produces this
> warning? Is it a normal consequence of this kind of
> computation (and can it therefore be disregarded), or does it
> suggest that something is wrong with the data?
>
> I know this has been asked of the list before, but the two
> posts I found dealing with this seem to have gone unanswered.
> I'd be very grateful for any assistance at all.
>
> Many thanks,
> Shelly Fitzgerald
>
> _________________________________________
> Shelly Fitzgerald
> Wisconsin Comprehensive Memory Program
> 2500 Overlook Terrace
> Madison, WI 53705
> (608) 256-1901 ext. 12912 (tel)
> (608) 280-7165 (fax)
> www.brainmap.wisc.edu
>
|