Print

---

Hi,
    Well spotted:) This is now fixed in our internal code - and will be included in the next release/patch.

Many Regards

Matthew
> Hi,
> 
> while testing an adaptor to allow using FSL in Condor pools, I found,
> what I believe is, a small bug in FEAT.
> 
> diff --git a/src/feat5/feat b/src/feat5/feat
> index 7953ab7..ea0cc5a 100755
> --- a/src/feat5/feat
> +++ b/src/feat5/feat
> @@ -371,7 +371,7 @@ if { $fmri(featwatcher_yn) } {
> 	    set howlong 300
> 	}
> 	    if { [ info exists fmri(featModelOpts) ] } {
> -		    set jobHalt {-h featModelID}
> +		    set jobHalt {-h $featModelID}
> 		}
> 	set prestatsID [ fsl:exec "${FSLDIR}/bin/feat ${fsfroot}.fsf -D $FD -I $session -prestats" -b $howlong $jobHalt -N feat2_pre -l logs ]
>     }
> 
> In SGE this doesn't seem to have a negative effect (at least when one
> only runs a single FEAT pipeline per user), but Condor is more picky
> regarding job ids.
> 
> I also modified the FEEDS script to survive submitting jobs to a cluster
> and still be able to compute the error values (by simply waiting for the
> result to be available).
> 
> --- a/RUN
> +++ b/RUN
> @@ -297,6 +297,9 @@
> # run FEAT
> fsl:exec "${FSLDIR}/bin/feat ${FEEDSDIR}/results/design.fsf"
> 
> +# wait for FEAT to finish (in case of cluster job submission)
> +exec sh -c "while \[ ! -f ${FEEDSDIR}/results/fmri.feat/cluster_zfstat1_std.txt \]; do sleep 5; done"
> +
> puts "checking error on filtered functional data:"
> perror fmri.feat/filtered_func_data 0.1
> puts "checking error on raw Z stat images:"
> @@ -397,6 +400,9 @@
> fsl:exec "cp -r $FEEDSDIR/data/fdt_subj1 $FEEDSDIR/results"
> fsl:exec "unset FSLMACHINELIST; ${FSLDIR}/bin/bedpostx $FEEDSDIR/results/fdt_subj1 -n 1"
> 
> +# wait for FEAT to finish (in case of cluster job submission)
> +exec sh -c "while \[ ! -f ${FEEDSDIR}/results/fdt_subj1.bedpostX/merged_th1samples.nii.gz \]; do sleep 5; done"
> +
> puts "checking error on bedpost output:"
> perror fdt_subj1.bedpostX/dyads1 .005
> perror fdt_subj1.bedpostX/mean_f1samples .005
> 
> 
> With these two modifications and a 'qsub'-emulation FSL seems to run
> fine with Condor. If anybody is interested in trying it out, let me know.
> In any case, one of the next Debian package updates will come with
> out-of-the-box support for Condor. This should make it a little easier
> for people to move away from SGE.
> 
> 
> Michael
> 
> 
> -- 
> Michael Hanke
> http://mih.voxindeserto.de
>