Chris Markiewicz
Chris Markiewicz
@yarikoptic copying https://github.com/bids-standard/bids-specification/pull/2044#issuecomment-2652590366 here, where it's more directly on-topic: > > My understanding was your preference was to classify stimuli as a new dataset type and validate its contents separately?...
@yarikoptic If `stimuli/` contents are to be specified, then the way to do that in the current schema would be to treat `stimuli/` as a datatype. If that's unacceptable, then...
Making `stimuli/` a conditionally-validated directory would be a new schema/validator concept. The only way I can see to avoid it is to use a different directory name in your BEP....
The proposal is the new concept I was saying we could avoid if you use a different directory name.
Right now fitlins does not have a notion of how much memory each task takes, so the scheduler will not help much with memory constraints. You can limit simultaneous tasks...
Is this profile with smoothing or without? We can probably separate these jobs to keep memory lower. You can also try using uncompressed niftis to take advantage of memory mapping....
It's built-in to nibabel. That said, it's possible to write code that unnecessarily copies data into memory, so hard to say whether we'll see any improvements. I think they're depending...
I suspect it'll be the model fit, which needs to load everything in memory. One hack could be to set an upper bound on number of voxels and parcelate the...
Where's the relevant `nb.load(img).get_data()`?