[FSL] BEDPOSTX Revisited

[oesteban]

• Update the nipype interfaces BEDPOSTX and XFibres. BEDPOSTX has a new command line interface as of FSL 5.0, which replicates XFibres interface. Former interface is still available, and usable in nipype using OldBEDPOSTX. XFibres has some new adaptations.
• Backwards compatibility: old interfaces are kept as OldBEDPOSTX and OldXFibres.
• Update create_bedpostx_pipeline
• Created a bedpostx_parallel workflow, that makes a better slicing of data to enable parallel processing in small chunks (when processing diffusion data with many sampled directions, memory consumption explodes even for the slice-by-slice approach).
• Created two new miscelaneous interfaces (SplitROIs and MergeROIs) as helpers for parallel pipelinings. Also created a regresion test to check consistency between both.

[coveralls]

Coverage increased (+0.02%) when pulling 3f677eb on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage decreased (-0.01%) when pulling 0f1c179 on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage decreased (-0.02%) when pulling 449452f on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage decreased (-0.05%) when pulling dc1a4cd on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage decreased (-0.05%) when pulling 0de90cf on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage increased (+0.02%) when pulling 6c01665 on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[coveralls]

Coverage increased (+0.05%) when pulling f177e3f on oesteban:enh/UpdateBEDPOSTX into f626d94 on nipy:master.

[oesteban]

I've been executing the new parallel workflow with the HCP data and looks good, but we could benefit from somebody else taking a look on this PR before merging.

[chrisgorgo]

We are dealing with breaking backward compatibility here. Maybe it's better to add a deprecation warning to the existing interfaces that need updating, create aliases for them with version number in the name (instead of "Old"). This will allow users to use new interfaces (with explicit versioned name) now but will also give other users time to update their code. We will switch the aliases in the next release. What do you think @satra ?

This release
XFibers -> XFibers4
XFibers4 (deprecation warning)
XFibers5

Next release
XFibers -> XFibers5
XFibers4
XFibers5

[satra]

is there any chance we can do it like the FSLGLMInput/OutputSpec?

basically it uses FSL's version to determine what parameters are available. also this resulted in deprecation of ContrastMgr as of FSL 5.0.7.

[satra]

also for BEDPOSTX it would be good to have support for BEDPOSTX_GPU which is simply a change of command.

class BEDPOSTXGPU(BEDPOSTX): _cmd = 'bedpostx_gpu' 

[oesteban]

Ok, I'll have news soon about all these suggestions 👍

[oesteban]

@chrisfilo my last commit should be OK with the new naming convention.

@satra, I've added support to the gpu version, and also implemented the interface choice based on FSL version. Still, I'm unsure about the actual implementation I've done (see this last part: oesteban@2afca1b#diff-367197de1c2db8cdd0dc656b3c96476dR1153)

[satra]

@oesteban - i'll review it later today!

[coveralls]

Coverage increased (+0.07%) when pulling 6e8c5ae on oesteban:enh/UpdateBEDPOSTX into 9fef6d4 on nipy:master.

[satra]

i'm going through this now. @oesteban - will send a pull-request shortly

[satra]

@oesteban - sent a pull request to your repo

[satra]

the travis error in this is due to the fact when there is no fsl, the auto test i generated for fsl 5.0.7 fails. the test should be skipped if there is no fsl. let me see if there is a quick fix for this.

[satra]

@oesteban - just sent a PR to your branch - should help with the tests.

[satra]

@oesteban - for this PR, could you please merge with upstream master, make specs, commit and push? for some reason a couple of files are being included that shouldn't have been.

[oesteban]

@satra: true, both test_auto_{BEDPOSTX,XFibres}.py were removed with the commit.

[coveralls]

Coverage increased (+0.04%) when pulling 13c7cfa on oesteban:enh/UpdateBEDPOSTX into 6f50ee1 on nipy:master.

[satra]

@oesteban - i'm going to merge this. can you continue testing this with your workflows?

[oesteban]

@satra sure! I'm processing a batch of cases from the HCP