TVB-2719 Add local_gdist_matrix tests and fix Bug 21

ayan-b commented 4 years ago

Closes #21

ayan-b commented 4 years ago

This PR is on top of #37.

liadomide commented 4 years ago

This PR is on top of #37.

As I just merged #37 I suggest a rebase with trunk

ayan-b commented 4 years ago

The first part of issue #21 is done. The matrix will not be exactly symmetric owing to precision errors involving double arithmetics. The same is done in the tests: assert (abs(distances - distances.T) > 1e-6).nnz == 0.
However, the second part is kind of tricky. In fact, the author has said the result is unreliable (again possibly due to precision errors). https://github.com/the-virtual-brain/tvb-gdist/blob/8c2bf68f2d4a821ebedbf2e858bf7dd3b76b9be2/geodesic_library/geodesic_algorithm_exact.h#L1284-L1287 One way to fix this is to assign distance[x][y] = distances[y][x] while calculating x -> y only once, in this way, the number of calculations will reduce by a factor by 2. But, I am not really sure if this is desirable.

the-virtual-brain / tvb-gdist