Hi
On 29 Jul 2008, at 11:09, Sascha Purmann wrote:
> Hi Christian, Dear FSL list:
>
> it would be strange if it were a memory problem, because I have the
> problems on three different computers, one running debian, one running
> Ubuntu 7, and the other running Ubuntu 8 64bit. I have tested RAM with
> memtest-86 on one computer, but will also run it on the others just to
> rule out the possibility that the errors result from bad memory.
>
> 1.) But why should there be a difference between case 2 and case 6
> (same amount of data)?
>
Same amount of data but internal data handling is very different -
case 2 compared to case 6 needs about twice as much memory *at peak*
> 2.) When you say 4GB RAM should be fine, would it work on a set of
> data from 12 subjects, with 2 session (880 epis) each?
>
Yes, but to make sure you can always cut down on the number of vols to
see if that improves - or set the resampling resolution to something
higher (e.g. 5mm) to reduce load
> 3.) In the end I want to have task related ICs and examine if they
> differ between two conditions (within-subject). Therefore, tensor ICA
> on all data is not necessarily what I need. How could I do that based
> on ICAs performed on single sessions? Any ideas?
>
Well, you still would want to check if that difference is consistent
across the population, so would need to combine across subjects at
some point. You could run single run ICAs, identify the main task-
related component and then calculate the mean signal difference
between conditions that way... just a thought
cheers
Christian
> Cheers, Sascha
>
> On Mon, Jul 28, 2008 at 11:11 PM, Christian F. Beckmann
> <[log in to unmask]> wrote:
>> Hi,
>>
>> is it possible that your system has some dodgy RAM? if run 6 2orks
>> but run 2
>> doesn't than this points to some memory related issue - but with
>> 4GB all
>> should be fine. Can you run some diagnostics?
>> If melodic crashes then yes, the report directory will be incomplete.
>> cheers
>> christian
>>
>>
>> On 25 Jul 2008, at 12:55, Sascha Purmann wrote:
>>
>>> [...]
>>> What I did so far for testing (all on the fsl course data;
>>> /fsl_course_data/melodic/tensor):
>>>
>>> 1. run tensor ICA with functional runs of all four subjects ->
>>> failed
>>> 2. run tensor ICA with functional runs of two subjects -> failed
>>> 3. split functional runs of subject 1 into two parts and run
>>> tensor ICA an
>>> these -> worked
>>> 4. split functional runs of subject 2 into two parts and run
>>> tensor ICA an
>>> these -> worked
>>>
>>> 5. run tensor ICA on part 1 of subject 1 and part 1 of subject 2 -
>>> > worked
>>> 6. run tensor ICA on all 4 parts
>>> (part1/subj1,part2/subj1,part1/subj2,part2/subj2) -> worked
>>>
>>> When tensor ICA failed, then with the error given in my previous
>>> message,
>>> which also means it does not give a full html-report (report/
>>> 00index.html
>>> does not show any components, TICA modes or PCA estimates).
>>>
>>> Especially 6 and 2 are interesting, because it's the same data but
>>> in 6
>>> split into pieces.
|