Hi Mark,
Thanks for looking into my issues!
I ran some simulations yesterday with the MNI template instead of out
patient-specific tissue priors and found that it worked for small and
large nproc. Investigating further I realized that the MNI template
has a different convention and orientation than our input, and was
able to fix it using fslorient and fslswapdim. I ran another
simulation overnight and all is well. It seems like POSSUM only takes
input in the "radiological" convention and in a specific orientation
("RPI")? Of course I probably should have realized this earlier, but
it might still be useful for beginners to POSSUM to include the
orientation requirements in the fsl/possum/input website. (Just a
suggestion!)
Thanks again for your time!
- Chintan
On Thu, Sep 15, 2011 at 9:36 AM, Mark Jenkinson <[log in to unmask]> wrote:
>
> Dear Chintan,
>
> You can ignore the extra job, or better still make the change from -le to -lt.
> This is already in our internal code but unfortunately has not yet made it to a patch.
>
> As for the problems you are seeing, have you checked that all the individual
> processes are completing correctly and without errors? We see this sort of
> behaviour when individual processes fail to generate correct output (or any
> output). If this is not the case then please give us some more details about
> the set up of your possum job (the saved GUI settings, along with the fslhd
> output from any customised input images, would be most useful).
>
> All the best,
> Mark
>
>
> On 14 Sep 2011, at 17:48, Chintan Shah wrote:
>
> > Hi all,
> > I am having a few problems with running possum in a cluster environment that I was hoping I could get some help with.
> > First, I noticed that possumX was creating an extra job (i.e. 101 jobs for 100 procs) and was wondering whether this was a problem or not. I found the post below (from several years ago) in which I noticed that the line
> > for ((procnum=0; procnum < $nproc; procnum++)) ; do
> > was replaced with
> > procnum=0
> > while [ $procnum -le $nproc ]
> > It looks like -le was used instead of -lt and so it produces an extra job. I don't think this makes a difference, as looking into possum_sum it seems like it only uses partial signal files from 0 to nproc-1. I just wanted to point it out and also ask if there are any known issues related to it, since I'm not sure exactly how possum splits up jobs.
> >
> > My main issue is also with regard to how possum splits up jobs. I'm slowly scaling up to run a larger simulation with motion on our cluster, but I first tried a small sim with just a single volume. When I run it with 8 processors it works fine. However when I run the exact same setup with 40 processors, it produces strange, almost rectangular output (in the axial view), which becomes even more distorted with nproc=100. I've attached some images to help describe what I'm talking about. Has anyone else run into a similar issue with changing the number of processors? or does anyone have a suggestion? Any help/comments would be appreciated!!
> >
> > Thanks,
> > Chintan
> > <axial_8.png><axial_40.png>
|