Possible spurious failure in Pysmo kriging tests

Question

Possible spurious failure in Pysmo kriging tests

Closed this issue a year ago · 3 comments

This occurred in https://github.com/IDAES/idaes-pse/actions/runs/6041051733 for a single job run and seems unrelated to #1253.

Answer 1 · 2023-09-07T14:23:06.000Z

The Pysomo tests are known to have an issue with non-deterministic behavior. Specifically, I think it is because the sampling steps are not seeded and thus we get a new set of sample points each time we run the tests and occasionally this results in an infeasible solution.

Answer 2 · 2023-09-07T18:20:06.000Z

@OOAmusat would this be easy/possible to fix as well?

Answer 3 · 2023-09-07T22:21:40.000Z

@lbianchi-lbl @ksbeattie This is due to random number generation with NumPy/SciPy in the initialization and solution steps.

I have attempted to fix the issue by fixing the numpy random seed generator for this test (PR #1264), but since it is a random failure there is no way to know if the fix works.