Clean discrete surfaces elastic metric #1986
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1986 +/- ##
==========================================
- Coverage 92.10% 88.82% -3.27%
==========================================
Files 158 147 -11
Lines 14442 13711 -731
==========================================
- Hits 13300 12178 -1122
- Misses 1142 1533 +391
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
| @@ -514,26 +514,23 @@ def __init__(self, space, a0=1.0, a1=1.0, b1=1.0, c1=1.0, d1=1.0, a2=1.0): | |||
| self.d1 = d1 | |||
| self.a2 = a2 | |||
|
|
|||
| self.exp_solver = DiscreteSurfacesExpSolver(space, n_steps=10) | |||
| if not gs.__name__.endswith("numpy"): | |||
There was a problem hiding this comment.
Isn't there a less hacky way to test the backend?
There was a problem hiding this comment.
for the name no... probably something to add.
|
Changing the division to a multiplication in |
746fc4d to
3fc6bd8
Compare
…this comes from the fact laplacian is not normalized by vertex areas) (brings code back to previous state)
3fc6bd8 to
c8723b7
Compare
|
After looking carefully to some equations, I believe Good to merge from my side. |
This PR cleans the discrete surfaces elastic metric:
einsums are replaced by proper backend functions
docstrings are fixed and simplified
default solvers are only instantiated if an autodiff-based backend is used (this way we can compute inner products with numpy)
a potential bug is solved in
_inner_prod_a2(@emmanuel-hartman, could you please confirm we should multiply by vertex areas instead of divide? this comes from the original implementation)