-
Notifications
You must be signed in to change notification settings - Fork 301
Memory errors #766
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Comments
|
|
|
|
|
|
|
|
|
|
|
Some memory issues still persist:
This memory error seems totally related to one on |
Worth considering packaging |
Would it be just plug and play? Or we need to revise nibabel's loads? |
Unfortunately, it had to be kept from being fully plug and play, because it clashed with an existing behavior of closing and releasing filehandles, to keep the OS from exhausting them, which apparently Gael ran into some years back. To enable it (if available) I would at to the top of import nibabel
nibabel.arrayproxy.KEEP_FILE_OPEN_DEFAULT = 'auto' I'm not sure how well that context will be passed when running function nodes, but it should work fine for anything that doesn't have to be pickled. We may want to do some tests. |
Thanks a lot. Before going down that road, I think I've found the biggest issue here. For ds000232/ sub-05_ses-04_task-main_run-01 we have the following:
I think we should calculate confounds (dvars, signals) on the original matrix size and possibly try float16? The aftermath is related to https://neurostars.org/t/why-are-functional-data-files-output-by-fmriprep-in-t1w-space-so-much-larger/796/4?u=oesteban . We should definitely keep the original pixel resolution in T1w space. I would unfold this memory issue in two:
WDYT @chrisfilo, @effigies ? |
Maybe a sensible approach for cropping BOLD series is to inflate the mask (1-2cm in all directions?), and crop that. |
Ok, yes, I see that the resolution is the same. Only the extent of the volume is extended to that of the reference (T1w) image. Well, that answers to the second question, when writing output files. I'll first address that one, because we probably don't need to worry about the first one (calculate confounds on the original space) since we are applying all transforms at once. It should be equivalent. |
@effigies FYI cropping is implemented in nilearn. One downside of this approach is that each output file might have different field of view (dimensions in voxels) - across participants that is. Not sure if it matters for T1w space outputs though. |
This PR implements this suggestion nipreps#766 (comment) If we find this useful, it is worth considering to add a new nipype config that leverages indexed_gzip with nibabel. Hopefully, this is the last piece of several PRs to fix nipreps#766
Uh oh!
There was an error while loading. Please reload this page.
The node that crashed the most times was
mri_coreg
followed byautorecon
. Worth looking into the requirements these nodes need.The text was updated successfully, but these errors were encountered: