MAINT: Permit recent nilearns

[effigies]

Summary

Fixes #2791.

List of changes proposed in this PR (pull-request)

Acknowledgment

• (Mandatory) I acknowledge that this contribution will be available under the Apache 2 license.

[codecov-io]

Codecov Report

Merging #2841 into master will increase coverage by 17.11%.
The diff coverage is 58.57%.

@@             Coverage Diff             @@
##           master    #2841       +/-   ##
===========================================
+ Coverage    50.4%   67.52%   +17.11%     
===========================================
  Files         126      299      +173     
  Lines       25304    39487    +14183     
  Branches        0     5220     +5220     
===========================================
+ Hits        12755    26664    +13909     
+ Misses      12549    12112      -437     
- Partials        0      711      +711

Flag	Coverage Δ
#smoketests	53.02% <32.82%> (+2.61%)	⬆️
#unittests	64.75% <51.97%> (?)

Impacted Files	Coverage Δ
nipype/interfaces/ants/base.py	60% <ø> (+11.06%)	⬆️
nipype/interfaces/ants/__init__.py	100% <ø> (ø)	⬆️
nipype/interfaces/base/core.py	86.96% <ø> (+20.36%)	⬆️
nipype/interfaces/afni/preprocess.py	82.02% <ø> (ø)
nipype/interfaces/afni/svm.py	97.61% <ø> (ø)
nipype/interfaces/afni/utils.py	81.86% <ø> (ø)
nipype/interfaces/ants/resampling.py	83.69% <ø> (+36.92%)	⬆️
nipype/interfaces/base/specs.py	93.33% <ø> (+9.75%)	⬆️
nipype/interfaces/ants/utils.py	88.28% <ø> (+3.46%)	⬆️
nipype/interfaces/ants/segmentation.py	74.39% <ø> (+26.57%)	⬆️
... and 340 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fd52a30...8c368c9. Read the comment docs.

[effigies]

nilearn 0.4.2 to 0.5.0 was a pretty huge change. While they had two pre-releases, it looks like our nilearn tests don't run on Travis, which is where we do our pre-release tests.

So I'm pinning to the pre-releases, to help narrow down the window in which the change happened. It's unclear whether this is a nilearn bug.

[effigies]

I've decided this is a nilearn bug: nilearn/nilearn#1891, introduced in nilearn/nilearn#1711.

We can also try to accommodate this version, by making sure that we pass in appropriate dtypes, but I think it's more appropriately handled on their end.

[effigies]

Last two commits verify the source of the failure. (I cancelled all but the pytest jobs.)

[satra]

0.5.1 was released 7 days ago, perhaps this can now be re-tested and merged?

[effigies]

Rebased. Doubt it'll work.

[effigies]

Nope. They did some weird stuff and were unwilling to revert, and I haven't found time to dig through their code and figure out how to restore functionality.

[effigies]

@satra Good to go?