FIX: Limit NaN imputation and use mean non-zero value

[effigies]

• Limit imputations to FD and DVARS variants
• Impute mean non-zero value
• Warn in report on imputation
• Incorporate Fix #216 bids-standard/pybids#217

Fixes #56.

[adelavega]

👌

[adelavega]

My PR is merged. Oh by the way, maybe it's not a bad idea to warn users (or throw an exception) if n/as are found in non-imputable variables?

[effigies]

Actually, do we even want to by-default impute any but the expected n/as for the first volume? It seems to me that anything else should go through a transform or raise an error.

And to complicate matters further, non-steady-state volumes will push down the "first" volume.

[adelavega]

I agree. I just think a nice exception would be nice to have.

[effigies]

And to complicate matters further, non-steady-state volumes will push down the "first" volume.

Looks like fMRIPrep does not remove non-steady-state volumes before calculating FD/DVARS, so we don't actually need to account for that. (The nistats interface will need to zero them out, but we should already be doing that.)

          CSF  WhiteMatter  GlobalSignal   stdDVARS  non-stdDVARS  \
0  608.277697    84.968374    289.321942        NaN           NaN   
1  -36.020893    -1.145165     -8.573609  10.742792    365.722839   

   vx-wisestdDVARS  FramewiseDisplacement  tCompCor00  tCompCor01  tCompCor02  \
0              NaN                    NaN    0.000000    0.000000    0.000000   
1        11.746696               0.495028   -0.013686    0.028413    0.012123   

   ...   Cosine03  Cosine04  Cosine05  NonSteadyStateOutlier00         X  \
0  ...    0.00000  0.000000  0.000000                      1.0 -0.006812   
1  ...    0.10448  0.104445  0.104403                      0.0  0.000000   

         Y        Z      RotX  RotY  RotZ  
0  0.00000  0.00000  0.000000  -0.0   0.0  
1 -0.03263  0.37969  0.001518  -0.0   0.0  

[2 rows x 32 columns]

[effigies]

Okay. Modulo some prettifying CSS, this is working for me. @adelavega do you want to give it a shot?

[adelavega]

can you do a release/push to docker hub? it'll make it easier for me to test

[effigies]

So how does it actually get run for you? Do you call docker from neuroscout?

[adelavega]

I extend your build with a Dockerfile: https://github.com/neuroscout/neuroscout-cli/blob/master/Dockerfile

(just realized I may be re-installing fitlins unnecessarily)

[effigies]

0.0.5 is on Docker Hub.

[adelavega]

Cheers