forum.alglib.net
http://forum.alglib.net/

mindf_gdemo_auto stopping criterion
http://forum.alglib.net/viewtopic.php?f=2&t=4592
Page 1 of 1

Author:  Maxwell [ Thu Feb 15, 2024 8:49 am ]
Post subject:  mindf_gdemo_auto stopping criterion

Hello,

I have started to use the Generalized Differential Evolution MultiObjective solver in single-objective mode, leveraging the adaptative choice of the DE parameters.
While the solver works nicely and yields the expected results, I did not find a way to change the desired tolerance after which the solver is stopped. The only option I found in the documentation is the maximum number of iterations, which is not sufficient for my study case.
Is there a way to set a threshold on the relative change of the solution vector ? In other words, if the optimal solution does not change by more than X%, then the solver stops.

Thank you for your help.

Author:  Sergey.Bochkanov [ Sat Feb 17, 2024 9:28 pm ]
Post subject:  Re: mindf_gdemo_auto stopping criterion

Hi!

Presently, there is only iterations count condition. Remember that multiheuristics have very erratic convergence, they can stagnate for a long time, they can steadily make tiny steps, etc, etc.

GDEMO was introduced in the last release, it was mature enough to be used. But presently we still test it on synthetic and real life problems and accumulate experience with the solver behavior. Presently I am not ready to take the risk of promising that some convergence condition which takes something like EPSILON will find a solution with accuracy even vaguely related to EPSILON.

Promising is easy, but making that promise work is not that easy :)

Page 1 of 1 All times are UTC
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/