Hi!

We have seen some improvements when including bandpass filtering on task data as well as resting state data. With a TR of 0.901 s we figured to basically use a low pass filter by setting the first argument to a very low number (0.001) and the second argument to 1 / (TR *3 ) = 0.37. I.e.:

-regress_bandpass 0.001 0.37

(for a TR of 2 this would be 0.001 0.17).

What I have realized is that this setting causes the voxel.loop segment (3dDeconvolve) and GSLQ loop (when using anaticor) to be extremely slow. Basically only changing one increment every other minute. I know the X-matrix becomes enourmous and very compliated when including bandpass fitlering since the frequencies filtered are hanled in the regression.

Is this expected or are we doing someting wrong? (It's almost as slow both on network mounted drive and on local SSD disk). This is two runs with a total of 1000 dynamics (500/run). After "Wrote matrix values to file X.nocensor.xmat.1D" it takes a very long time (matrix setup tine = 287.31 s.) then the even slower voxel loop starts with a couple of minutes between each tick of 0123456789.12 etc..

Edit: All cpu cores are on 100 % during the voxel loop (matrix setup is just single core 100 %)

Thanks!

Robin



Edited 2 time(s). Last edit at 06/26/2017 11:09AM by Robin.