Dear FSL developers and users,
I have noticed that when you run a higher-level FEAT analysis, the step of the lower-level feat directories registrations (featregapply; part of the "Higher-level input files preparation"; name of job: feat2_pre) takes quite a lot of computation time when registration is run for the first time. The way it works now is that the featregapply command is run sequentially, i.e. after each other, for each of the input feat folders. Is it possible to implement a parallelized version of the featregapply step? If these jobs are run as an arrayjob, this step will save a lot of time for users using a computer cluster, especially if there are a high number of input files.
Of course users can just run featregapply manually before running higher level FEAT. Feat will then skip the registration if it was already run. Here is a handy oneliner to parallelize this step on the cluster.
ls -d1 /data/foldercontainingmyfeatfiles/subject???.feat | xargs -I TODO fsl_sub featregapply TODO
Best,
Henk van Steenbergen
|