-
Notifications
You must be signed in to change notification settings - Fork 158
Description
Describe the Issue
When I follow the instructions on the CONTRIBUTING.md. I found that test91, test MaxIB, failed. The log is shown as below.
Describe the solution
The current max_inscribed_ball() function in max_inscribed_ball.hpp does not normalize the input matrix. After I added lines to normalize each row, the function produced a valid result and passed the test.
Describe alternatives
I noticed that test91 calls ComputeInnerBall() to calculate the correct MaxIB to test max_inscribed_ball(). While ComputeInnerBall() calls max_inscribed_ball(). Therefore, the test is using the function to test itself, so it might not be reliable.
Besides, I wonder if the back-up method Lpsolve in ComputeInnerBall() is announced as unstable at high dimensions?(https://github.com/GeomScale/gsoc25/wiki/Exclude-Lpsolve) Currently, the test is at low dimension, but I can try to run it at high dimensions and replace it with other packages like Gurobi to see if they can be more stable.
Additional context
