FIX+ENH: refactor area weighted surface distance + fields operations

[oesteban]

• Fix area-weighted path
• Refactor former code to be more robust (now using numpy when possible)
• Add zero-distance regression test
• Add translation-distance regression tests
• Add area-weighting regression tests

[oesteban]

This PR is ready for code-review :)

[satra]

what if these two files are spheres? is there a way to weight them differently, for example with curvature information or some such?

[oesteban]

You're right, I did not observe that detail, but I could add a new weighting keyword that reads the vertex-wise properties of the vtk file, and definitely curvature may be a potential weight. I'll look at this ASAP.

[satra]

also as a side note, i wanted to write compy (a python computational geometry library) a while back that would allow us to do such operations easily (perhaps with vtk this is not longer necessary, or something already exists that i haven't followed),

i implemented a set of computational geometry algorithms (on triangular meshes) a long while back in matlab. they should be relatively straightforward to convert to python + cython.

https://github.com/satra/SLT/tree/master/SurfTools

if you are interested we can start compy (if no such other library exists) and chip away at this.

[oesteban]

Hi @satra, I definitely agree on that computational geometry algorithms would be much better developed and maintained as a package different from nipype. The algorithms you had implemented in matlab look as a good starting point, but also we could benefit a lot from vtk. From this side, everythings looks good to me in favor of compy.

However, I am currently struggling to finish up my PhD. For now, I can only contribute with stuff I really need for the results of my thesis... Thus, compy would start with a very particular and limited choice of algorithms from my side. What do you think?

[coveralls]

Coverage decreased (-0.16%) to 70.58% when pulling d0f0297 on oesteban:fix/AreaWeighedP2PDistance into 453bcf9 on nipy:master.