enable custom cluster based thresholding for non-volumic dataset

[bpinsard]

modified the GroupClusterThreshold to enable clustering using custom neighborhood definition (as a sparse matrix of connectivity) that can process surface based dataset for example.

from an array tris defining the triangles of the surface, connectivity can be obtained with:

scipy.sparse.coo_matrix((
            np.ones(3*tris.shape[0]),
            (np.hstack([tris[:,:2].T.ravel(),tris[:,1]]),
             np.hstack([tris[:,1:].T.ravel(),tris[:,2]]))))

I prefer to provide neighborhood than do something surface-specific as I have dataset mixing surfaces and ROIs (HCP-style).

Runs quite fast (comparable to the volume based).

Testing needed, but wanted to have feedback on the structure.

[yarikoptic]

Please have a look at #395 which wip to make it work with arbitrary queryengines and extended set of measures... Would be nice to join efforts

[codecov-io]

Codecov Report

Merging #453 into master will decrease coverage by 0.03%.
The diff coverage is 30%.

@@            Coverage Diff             @@
##           master     #453      +/-   ##
==========================================
- Coverage   76.66%   76.63%   -0.04%     
==========================================
  Files         364      364              
  Lines       41341    41366      +25     
  Branches     6667     6672       +5     
==========================================
+ Hits        31695    31700       +5     
- Misses       7751     7770      +19     
- Partials     1895     1896       +1

Impacted Files	Coverage Δ
mvpa2/algorithms/group_clusterthr.py	88.23% <30%> (-10.09%)	⬇️
mvpa2/tests/test_usecases.py	99.28% <0%> (+0.35%)	⬆️

---

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 56d8861...85da1df. Read the comment docs.

[coveralls]

Coverage decreased (-0.04%) to 79.682% when pulling 8f0ab92 on bpinsard:gct_custom_neigh into e22122c on PyMVPA:master.

[bpinsard]

I didn't see that it was WIP, I will follow your pull request. Thanks!

[coveralls]

Coverage decreased (-0.03%) to 79.911% when pulling 2ebaa84 on bpinsard:gct_custom_neigh into 1d11560 on PyMVPA:master.

[coveralls]

Coverage decreased (-0.03%) to 79.982% when pulling 64d8124 on bpinsard:gct_custom_neigh into b036f96 on PyMVPA:master.

[coveralls]

Coverage decreased (-0.05%) to 81.154% when pulling 6b9eb94 on bpinsard:gct_custom_neigh into 7e9020b on PyMVPA:master.

[coveralls]

Coverage decreased (-0.05%) to 81.154% when pulling 6b9eb94 on bpinsard:gct_custom_neigh into 7e9020b on PyMVPA:master.

[coveralls]

Coverage decreased (-0.04%) to 81.161% when pulling 6b9eb94 on bpinsard:gct_custom_neigh into 7e9020b on PyMVPA:master.

[coveralls]

Coverage decreased (-1.5%) to 79.652% when pulling f8d1b86 on bpinsard:gct_custom_neigh into 7e9020b on PyMVPA:master.

[coveralls]

Coverage decreased (-0.2%) to 79.859% when pulling 85da1df on bpinsard:gct_custom_neigh into 56d8861 on PyMVPA:master.