DLASCL error?

[Eric W]

Hi,

    Just did a complete run of CPAC on the KKI Kirby 21 dataset (42 subjects) and experienced the following error on 5 of the 42 subjects:

"On entry to DLASCL parameter number 4 had an illegal value"

I've enclosed the 5 crash files produced by CPAC as well for your records. It seems like it is not CPAC specific as it has been noted previously for data fitting (see here). I've enclosed all the crash files, so hopefully that will help you guys. Let me know if you have any questions about versions of different libraries on our machines. 

@jovo

[Cameron Craddock]

This is a problem that we have fixed in 0.3.9, Replace  U, S, Vh = np.linalg.svd(np.dot(Yc, Yc.T))  on line 14 in calc_compcor_components with 

U, S, Vh = np.linalg.svd(Yc) 

These two should give the same result (eigne decomp of rows and columns of Yc) but I think that the revised code is less sensitive to matrix conditioning.

-cc

--
You received this message because you are subscribed to the Google Groups "cpax_forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cpax_forum+...@googlegroups.com.
To post to this group, send email to cpax_...@googlegroups.com.
Visit this group at http://groups.google.com/group/cpax_forum.
To view this discussion on the web visit https://groups.google.com/d/msgid/cpax_forum/d6d62e61-2743-4101-972c-3d3b7ca73c63%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Director of Imaging, Center for the Developing Brain, Child Mind Institute

Research Scientist VI, Nathan S. Kline Institute for Psychiatric Research

Co-Founder, The Neuro Bureau

Impact Story

Google Scholar Citations

GitHub

Share that Brain @ the International Neuroimaging Datasharing Initiative!

Join me 2014-10-18 @ Brainhack EDT!

[Eric W]

Attached are 9 more crash files from 2 new CPAC runs (first 5 are from a 42 subject run with ANTs for anat reg, second 4 are from the same 42 subject run with FSL for anat reg). Same issue after applying the solution suggested. Any other potential fixes? Note, this is the Kirby 21 42 subject dataset. 

[Eric W]

... and the second 4 crash files, from a run with FSL for anat reg. 

[Cameron Craddock]

In general this error indicates that you are trying to invert something that is non-invertible. I think it maybe that you have some voxels whose time courses have zero variance. Would you try adding this line and see if it fixes your problem? 

https://github.com/FCP-INDI/C-PAC/blob/0.3.9_development/CPAC/nuisance/utils.py#L10

--
You received this message because you are subscribed to the Google Groups "cpax_forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cpax_forum+...@googlegroups.com.
To post to this group, send email to cpax_...@googlegroups.com.
Visit this group at http://groups.google.com/group/cpax_forum.

To view this discussion on the web visit https://groups.google.com/d/msgid/cpax_forum/c8ef554a-762d-4690-88f4-30098193650f%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.