edickie / ciftify

The tools of the Human Connectome Project (HCP) adapted for working with non-HCP datasets
https://edickie.github.io/ciftify/
MIT License
117 stars 157 forks source link

Looking for help on ciftify install #184

Open Mcstevens opened 10 months ago

Mcstevens commented 10 months ago

After years of putting it off, I finally dove into some dense CIFTI analyses instead of just focusing on easier-to-use parcellated data. So I wanted to use the really useful-looking ciftify_peaktable utility to help summarize findings, but am running into problems. Specifically, I get an error:

AttributeError: 'Index' object has no attribute 'get_values' traced back to line 182 in run_ciftify_dlabel_report.

The msg seems to suggest there is no input coming from the dense CIFTI file. But that's not the case. Not only are there lots of significant vertices and voxels to work with in my result .dscalar.nii file, ciftify_peaktable produces the associated .dlabel.nii file... and I can visualize it in wb_view. So this seems to be a problem of a different nature.

The first thing I tried to do to troubleshoot was take a clean Linux system and put the ciftify package onto it via pip so it'd be fresh. But I ran into the same sorts of package deprecation and incompatibilities (e.g., nibabel, numpy versions, etc) noted in recent posts. I managed to get things seemingly (?) working by installing nibabel 4.0.0 specifically. But still, I run into this "get_values" error no matter what I do. I've even tried it several other Linux machines in our cluster where ciftify's image conversion routines have stably worked for a couple of years.

Does anyone know the nature of this error or can make helpful troubleshooting suggestions? Or, Erin have you or anyone been working at tweaking the older code so that it installs cleanly with current updated prerequisites? I ask the latter because I can't fully rule out that I might be doing something wrong with the primary analysis (e.g., which surfaces and areal files I'm feeding into it). That is, perhaps ciftify_peaktable would work fine but I'm feeding a wonky image into it... despite it looking entirely plausible to me when I visualize the results. It would be helpful to rule out the platform or the code's install on our machines as the possible culprits however.

Many thanks, Mike