Dear Johann,
With many sessions, the estimation stage of SPM99b will make session
specific F-contrasts in the xCon contrast definition structure. Each
contrast definition contains the corresponding partitioning of the design
matrix. So, for large designs with many sessions, you end up with a lot of
F-contrasts, each of which contains (much of) the design matrix. Thie
results in a huge xCon.mat file, which is what your wee SPARC-5 had trouble
saving.
As a temporary fix, you can stop SPM99b from generating these default
session specific F-contrasts by "hacking" the SPMcfg.mat file. At the end
of the fMRI design setup, choose to estimate "later". Then, type the
following in the Matlab command window:
>> clear
>> load SPMcfg
>> F_iX0 = F_iX0(1).iX0
>> %-If this generates an error then you don't have session
>> % specific F-contrasts
>> save SPMcfg
You can then proceed to estimate the model, using the "Estimate" button on
the SPM Menu window, and selecting the SPMcfg.mat file in the present
working directory.
The final release of SPM99 will ask whether session specific F-contrasts
should be generated, and stores the implied partitioning in a more
efficient manner (thanks to JB).
Hope this helps,
-andrew
At 15:20 05/11/99 +0100, Johann P.Kuhtz wrote:
| I consistently experienced a crash when estimating a specified
| model: a large design matrix for fMRI with spm99b, using matlab
| 5.3. The crash happens when saving results. In the design matrix I
| have 48 sessions (8 subjects, 6 sessions per person) with 104
| scans per session (total 4992 scans). Each session contains 13
| epochs: 6 epochs of activation and 7 epochs of rest (1 epoch is 40
| sec long, = 8 scans, TR=5s). SOA is fixed. I modeled 3 conditions
| of a motor paradigm, and 6 motion parameters as confounds (user
| specified regressors, entering the "realignment-params.txt" -files
| for each session with spm_load). I use a box-car convolved with
| hrf, and scaling for global normalization. Highpass filter was set
| at 480 sec cutoff freq.
|
| The messages in the matlab window are as follows:
|
| SPM 99b: spm_spm (v2.28.1.1)
| Initialising design space : ...done
| Output images : ...initialised
| Plane 34/34, plank 48/48 : ...done
| Design parameters : ...done
| Smoothness parameters : ...done
| Saving results : ...writing (so far all is ok)
|
| [0 (459 0)]: segv (11). code 4, pc 0xda8776c, vaddr 0xff65063ff8, alt-sp
| 0x13037a20
|
| ??? error while evaluating uicontrol Callback
| >> Unable to create crash dump log file
| ---------------------------------------------------------------
| Segmentation violation detected at Thu Nov 4 21:23:24 1999
| ---------------------------------------------------------------
|
| Configuration:
| Matlab version: 5.3.0.10183 (R11)
| Operating system: IRIX64 6.5 11051731 IP27
| Window system: Sun Microsystems, Inc. (3600), display
| skalleper.child.ks.se:0.0
| Current Visual: 0x22 (class 3, depth 8)
|
| Please follow these steps .... 1. Send us this crash report ... (now some
| instructions of matlab appear)
|
| Error in ==> /usr/local/spm99b/spm_spm.m
| On line 975 ==> save ('xCon.mat', 'xCon')
|
| Error in ==> /usr/local/spm99b/spm_fmri_spm_ui.m
| On line 551 ==> spm_spm(VY,xX,xM,F_iX0,Sess,xsDes)
|
| So far this crash report. I use a SunSparc station 5. When I check memory
| by typing "limit" I get the information
|
| cputime unlimited
| filesize unlimited
| datasize unlimited
| stacksize 2000000 kbytes
| coredumpsize 0 kbytes
| memoryuse 3750000 kbytes
| vmemoryuse unlimited
| descriptors 200
| threads 1024
|
| and by typing "swap -s" I get
| total: 0.00k allocated + 206.23 add'l reserved = 206.23m bytes used,
| 10.27g bytes available
|
| I would be grateful for any suggestion how to correct this problem.
+ - Dr Andrew Holmes mailto:[log in to unmask]
| Robertson Centre for Biostatistics ( ,) / _)( ,)
| Boyd Orr Building, University Ave., ) \( (_ ) ,\
| Glasgow. G12 8QQ Scotland, UK. (_)\_)\__)(___/
+ - http://www.rcb.gla.ac.uk/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|