Pain Points of Tedana Meta-Issue

[jbteves]

Summary

We'll list common points of frustration or confusion here ahead of our hackathon

Running List

• [ ] How to run

[emdupre]

• [x] How to run from the fMRIPrep working directory
• [ ] How to run from AFNI proc
• [ ] How to incorporate fieldmap corrections appropriately
• [ ] Memory issues in running tedana
• [ ] Changing the default PCA denoising ?

[smoia]

• [ ] Debug PCA denoising
• [ ] Have a look at the elbow code for its PCA application
• [ ] Get a general background of the ME maths related to the code

[dowdlelt]

• [ ] using tedana after/with other software (SPM comes to mind)
• [ ] degrees of freedom after tedana
• [ ] using tedana regressors seperately

[jbteves]

@handwerkerd anything missing?

[handwerkerd]

• [ ] How to figure out if denoising ran correctly (i.e. easy summary stats)
• [ ] How to figure out what can be run better, if the summary stats show an issue
• [ ] How to decide which options to use
• [ ] How to denoise across multiple runs during one session, across sessions/subjects
• [ ] What steps to do before vs after denoising (i.e motion & distortion corrections)
• [ ] Integrating recorded respiratory and physiological information into multi-echo denoising

[dowdlelt]

As post-tedana steps are developled (#344) I imagine they will add an additional, but user-desired, level of complexity so

• [ ] How to use knowledge (motion estimates, maps) for component refinement

[jbteves]

I'm actually going to break the mold and add a developer point of frustration:

• [ ] Running and modifying tests

[dowdlelt]

An additional one related to figures,

• [ ] check orientation of brain/acquisition when making multiplanar cuts - rotate to normal.

Sometimes with R-L acquisitions the brain is flipped on its side, coronal isn't coronal, etc - That just adds a bit of frustration (and possible confusion) when inspecting output. May be an easy fix...

[dowdlelt]

And thought of another thing: the errors that pop up during running (not due credit, we know about that...). maybe these are environment specific, but they always worry me...

• [ ] runtimes errors that cause confusion and concern

some examples:

/home/stone-ext4/logan/.local/lib/python3.6/site-packages/sklearn/externals/joblib/__init__.py:15:
 DeprecationWarning: sklearn.externals.joblib is deprecated in 0.21 and will be removed in 0.23. 
Please import this functionality directly from joblib, which can be installed with: pip install joblib. If 
this warning is raised when loading pickled models, you may need to re-serialize those models with 
scikit-learn 0.21+.


RuntimeWarning: divide by zero encountered in true_divide
  F_S0 = (alpha - SSE_S0) * (n_echos - 1) / (SSE_S0)

INFO:tedana.metrics.kundu_fit:Performing spatial clustering of components
INFO:tedana.selection.tedica:Performing ICA component selection with Kundu decision tree v2.5
/home/stone-ext4/logan/.local/lib/python3.6/site-packages/tedana/selection/tedica.py:244: RuntimeWarning: invalid value encountered in double_scalars
  np.min(comptable.loc[acc_prov, 'variance explained'])))

[jbteves]

@dowdlelt FYI we've discussed a little bit about duecredit specifically in the now-closed #196. @tsalo at some point added the note that it doesn't impact package functionality.

[dowdlelt]

yeah, duecredit doesn't bother me at all, because it has the note and such - it is the other warnings while the package is running that are a bit more confusing. If users see problems in output, have trouble running, etc - then it would be natural for them to assume that these problems could be coming from the issues they see during code execution.

[jbteves]

Ack, sorry, I misread the above. I agree that we should fix the other warnings.

[smoia]

• [ ] Add a check on volumes and TE input order

[tsalo]

This has been untouched for three years now. Can we update our pain points (I assume at least some have been addressed) and break this up into separate issues?

[tsalo]

I think we have either addressed or opened separate issues for each of the pain points brought up in this issue, so I'm going to close it.