We currently support 3 different minimization engines, all these 3 engines behaves differently when the fit does not converge:
DFO
DFO will throw a FitError with an appropriate error message:
LMFit
LMFit does not throw or warns, but its success flag is properly set to False
Bumps
Bumps does not throw or warns, AND its success flag is wrongly set to True because the fit works and doesn't throw anywhere (it JUST reaches max_iterations).
While not a bug, I also wish to have evaluations as a result in our Result object, so that it can easily be inspected for profiling without having to know its underlying name in our minimizer backends.
We currently support 3 different minimization engines, all these 3 engines behaves differently when the fit does not converge:
DFO
DFO will throw a
FitErrorwith an appropriate error message:LMFit
LMFit does not throw or warns, but its
successflag is properly set toFalseBumps
Bumps does not throw or warns, AND its
successflag is wrongly set toTruebecause the fit works and doesn't throw anywhere (it JUST reaches max_iterations).While not a bug, I also wish to have
evaluationsas a result in ourResultobject, so that it can easily be inspected for profiling without having to know its underlying name in our minimizer backends.