schlegelp
commented
2 years ago A first quick & dirty test yielded mixed results: sometimes the robust-laplacian is marginally faster, sometimes our own. The main problem is that the number of steps it takes to get below epsilon varies which in turn makes for a bad comparison: if three steps get you within 1% of the target epsilon it will run a fourth which will (a) take much longer and (b) overshoot the target.
The results from robust-laplacian look better though. Might be because of the mollify_factor?
Note to self: check out https://github.com/nmwsharp/robust-laplacians-py for the mesh contraction.