Dear Helmut,
I suggested changing stats.maxmem as it might help running the
estimation faster (esp for the FFX model) but sadly it won't fix your
'out of memory' problem. [Note that until SPM8, data from at most one
slice were used at a time for model estimation while with SPM12, this
restriction is lifted and stats.maxmem directly controls how much data
can be used at the same time, so a reasonable number has to be entered
there].
I ran something similar than you a few weeks ago (with a slightly larger
dataset) - the 12GB of RAM of my computer were not enough and I went and
used a dedicated machine instead (differences in results were not worth
the effort, which is a good news).
If you model include realignment parameters, you might want to adjust
the variables iX1 and iX0 that define the partition of first level
design matrices into effects of interest and confounds or nuisance
variables (spm_mfx.m at l.122-123): the indices of the columns
corresponding to the movement parameters should go in iX0 instead of iX1
as they are by default.
Best regards,
Guillaume.
On 04/03/14 12:33, H. Nebl wrote:
> Dear Torben, dear Guillaume,
>
>
> well yes, it's a standard PC and 4 (?) years old, that's true of course. At least it is still better than the usual stuff we get/have to take from our data center ;-) I've increased the defaults.stats.maxmem a while ago, but I have to check the exact value when back on that computer.
>
> Running models on a subset of subjects (12 subjects, 1 session each) would actually be of no use (basically I want to compare the results with those from the standard two-stage procedure). Realignment parameters were included.
>
> I was just surprised a little that the fixed-effects model based on the 12 data sets worked well, whereas the mixed-effects model does not. It should be the same amount of data, shouldn't it. Anyway, let's see whether adjusting defaults.stats.maxmem helps.
>
>
> Best,
>
> Helmut
>
--
Guillaume Flandin, PhD
Wellcome Trust Centre for Neuroimaging
University College London
12 Queen Square
London WC1N 3BG
|