Better Handling failed points
Issue #41
resolved
Currently, failed points are just used to prevent future evaluations. However in cases like Kriging, it becomes important how these are handled for future iterations.
One way is to just use the interpolant (current mean) as the function value for failed points. This will not affect the mean but changes the variance of the surrogate.
Comments (4)
-
reporter
-
reporter
commit #5447807 partially completes this bug. More rigorous tests and MLE optimization have to be checked.
-
reporter
commit #755ca40 has the changes for MLE optimization. It only considers the successful points for MLE and not others.
-
reporter
- changed status to resolved
7368bfd should resolve this issue... tests are showing good results with MLE.
- Log in to comment
MLE optimization has to be carefully done.