[ENH] New ComputeDVARS interface

[oesteban]

Computes DVARS, standardized DVARS and voxel-wise standardized DVARS.

Additionally, correspondingly plots are generated:

[coveralls]

Coverage decreased (-0.1%) to 72.128% when pulling 38229a0 on oesteban:enh/NewDVARSInterface into 0f9a620 on nipy:master.

[coveralls]

Coverage decreased (-0.1%) to 72.128% when pulling 38229a0 on oesteban:enh/NewDVARSInterface into 0f9a620 on nipy:master.

[coveralls]

Coverage decreased (-0.1%) to 72.128% when pulling 38229a0 on oesteban:enh/NewDVARSInterface into 0f9a620 on nipy:master.

[coveralls]

Coverage decreased (-0.1%) to 72.134% when pulling 3250176 on oesteban:enh/NewDVARSInterface into 0f9a620 on nipy:master.

[chrisgorgo]

Could you add a test on top of the autogenerated one?

[oesteban]

I'll see how, np.

@chrisfilo - One edit:
Maybe a cross-comparison vs. https://www2.warwick.ac.uk/fac/sci/statistics/staff/academic-research/nichols/scripts/fsl/DVARS.sh ?

Another edit:
http://czarrar.github.io/checking-on-dvars/

[satra]

merge with master

[chrisgorgo]

Comparison with @nicholst script sounds like a good idea.

[oesteban]

@chrisfilo: the regression test it a bit permissive (MSE < 0.05) because the robust estimation of the SD in the @nicholst's script and this implementation differ. For some reason the results from fslmaths and numpy are not exactly the same. Surprisingly the lag-1 coefficients are the same (using nitime and fsl), so I think we can go ahead with this implementation.

[chrisgorgo]

I think you meant @nicholst not @nicholsn ;)

[oesteban]

My apologies to the mistakenly mentioned user.

I had to add nitime to the requirements.txt. I guess that, as we move towards setuptools (#1605), we will need to add these semi-optional dependencies to the extras_require option of setup().

[satra]

@oesteban - yule-walker should be easy to implement within the code, instead of another dependency. there is a set of nitime interaces in nipype, but we don't add nitime to requirements.txt.

[oesteban]

You got it. For now I will skip the test if nitime is not found.

We could incorporate automated testing of dependencies into the Interfaces, using a class member like dependencies_.

[coveralls]

Coverage increased (+0.2%) to 72.468% when pulling 585c0fc on oesteban:enh/NewDVARSInterface into bd3a7ef on nipy:master.

[berleant]

Robust standard deviation sounds useful--we should consider moving it to another module so everyone can use it.

[oesteban]

Well, I'd say that this is a quick and dirty way of calculating the SD. I am +1 if we had a really robust SD computation algorithm though.

[berleant]

got it

[codecov-io]

Current coverage is 70.66% (diff: 68.21%)

Merging #1606 into master will increase coverage by 0.03%

@@             master      #1606   diff @@
==========================================
  Files          1016       1017     +1   
  Lines         50928      51076   +148   
  Methods           0          0          
  Messages          0          0          
  Branches       7238       7255    +17   
==========================================
+ Hits          35971      36094   +123   
- Misses        13894      13904    +10   
- Partials       1063       1078    +15   

Powered by Codecov. Last update 5c9a751...985d4e3

[oesteban]

I think this is ready to go