 |
 |
Hi Alexandra, I just tried running an existing cross-session model on slurm with automatic outlier deweighting and I also got some errors. I was able to narrow the error down to an issue generating res4d. There was only an issue with some of the copes. /projects/ics/software/fsl/5.0.8/bin/smoothest -d 4 -m mask -r stats/res4d > stats/smoothness Mask and Data (residuals/zstat) volumes MUST be the same size! The res4d only had 75 volumes, but zstat had 91, so I think the error came from the fslmerge step : /projects/ics/software/fsl/5.0.8/bin/fslmerge -z stats0000/res4d.nii.gz stats0000/res4d stats0001/res4d stats0018/res4d stats0019/res4d ...' I ran this with the fsl_sub defaults, and have had issues with fslmerge in our instances due to SLURM memory container limits. So as a next step I re-ran the model setting a flag in our fsl_sub to set the default number of cpus per job to 2 (i.e., the slurm '-c' option). This is our workaround for expanding the memory limits. So you may be experiencing a memory limit if your system has containers defined. Joe -- Joseph M. Orr, Ph.D. Assistant Professor Department of Psychology Texas A&M University College Station, TX On Fri, Nov 20, 2015 at 10:26 AM, Alexandra DiFeliceantonio < [log in to unmask]> wrote: > Hello Matthew, > I am using SLURM. > Through toggling around some parameters, I found the problem was with the > automatic outlier deweighting. With that option turned on SLURM doesn't > handle things properly. Any suggestions for altering the fsl_sub to fix > this would be greatly appreciated. > Best, > Alexandra > > On Fri, Nov 20, 2015 at 6:23 PM, Matthew Webster < > [log in to unmask]> wrote: > >> Hello, >> Are you running this in some kind of parallel environment >> (e.g. a compute cluster )? A number of slices have not completed, which can >> be due to a cluster queue timeout. >> Kind Regards >> Matthew >> >> > Dear Experts, >> > I using FEAT to analyze data from subjects with multiple runs. >> Everything is fine on the 2nd level and a cope*.nii.gz is created for each >> subject. For my Group level, I entered those cope*.nii.gz files as inputs. >> And then something strange happens. >> > I get this error >> > /daten/share/opt/fsl/fsl-5.0.8-cscor/bin/smoothest -d 16 -m mask -r >> stats/res4d > stats/smoothness >> > Mask and Data (residuals/zstat) volumes MUST be the same size! >> > >> > When I check res4d it has 49 slices. >> > >> > It seems to be originating from above: >> > /daten/share/opt/fsl/fsl-5.0.8-cscor/bin/fslmerge -z >> stats0000/res4d.nii.gz stats0000/res4d stats0001/res4d stats0002/res4d >> stats0003/res4d stats0004/res4d stats0005/res4d stats0006/res4d >> stats0007/res4d stats0008/res4d stats0009/res4d stats0010/res4d >> stats0011/res4d stats0012/res4d stats0013/res4d stats0014/res4d >> stats0015/res4d stats0016/res4d stats0059/res4d stats0060/res4d >> stats0061/res4d stats0062/res4d stats0063/res4d stats0064/res4d >> stats0065/res4d stats0066/res4d stats0067/res4d stats0068/res4d >> stats0069/res4d stats0070/res4d stats0071/res4d stats0072/res4d >> stats0073/res4d stats0074/res4d stats0075/res4d stats0076/res4d >> stats0077/res4d stats0078/res4d stats0079/res4d stats0080/res4d >> stats0081/res4d stats0082/res4d stats0083/res4d stats0084/res4d >> stats0085/res4d stats0086/res4d stats0087/res4d stats0088/res4d >> stats0089/res4d stats0090/res4d >> > >> > >> > Where it's creating a file with 49 slices instead of 91. >> > >> > Further up in the log I see this message for many of my slices (but not >> necessarily the ones that are missing): >> > WARNING: The passed in varcope file, tmpvarcope0032, contains voxels >> inside the mask with zero (or negative) values. These voxels will be >> excluded from the analysis. >> > >> > >> > The input copes all have 91 slices. Anyone know where this could be >> originating from? As part of FEATS clean up it seems to delete some of the >> files it would be useful at looking at to determine the problem. >> > Any help would be greatly appreciated. >> > Best, >> > Alexandra >> > > > > -- > Alexandra DiFeliceantonio, PhD > Post Doctoral Fellow, Max-Planck-Institute for Metabolism Research > Visiting Post Doctoral Fellow, John B Pierce Laboratory at Yale University > +49 221 4726631 > +1 203 562 9901 ext 209 > "It would be surprising indeed if the brain were organized into spatially > discrete units that conform to our abstract conceptualizations of behavior" > Eliot Valenstein > >