GlobalSearch - Does a non-default number of trial points stop the effect of the maxTime option?

5 views (last 30 days)
Hi,
I am using GlobalSearch() to find an optimum of a solution of odes in fmincon. The functions in fmincon have many input variables with upper and lower bounds each. That's why I would like to increase the number of trial points from the default value of 1000 to maybe 5000. With 1000 trial points, a minimum often can be found withut exceeding the maxTime (e.g. 1 h = 3600 s). But sometimes GlobalSearch exceeds the maxTime and is then stopped accordingly after the current fmincon run. That is desired behaviour.
When I change the number of Trial Points to 5000 with a maxTime of e.g. 1 h the maxTime option doesnt seem to stop GlobalSearch anymore, when one hour is exceeded.
Did I misunderstand the use of Trial Points or is this a bug?
Thanks in advance!

Answers (1)

Pratyush
Pratyush on 15 Apr 2024 at 12:41
Hi,
Increasing the number of trial points in MATLAB's GlobalSearch algorithm to 5000 seems to cause issues with the maxTime parameter not effectively stopping the process after the specified duration. This behavior could be due to a few reasons:
  1. Behavior of maxTime: GlobalSearch may allow a current fmincon run to complete even if maxTime is exceeded, leading to longer total runtimes.
  2. Impact of Trial Points: A higher number of trial points means more preliminary evaluations, which could consume a significant portion of the allotted time before the more intensive local optimizations begin.
Recommendations:
  • Use diagnostic settings ('Display', 'iter' or 'diagnose') to gain insights into the optimization process and time allocation.
  • Consider reducing the number of trial points or trying a different optimization algorithm like MultiStart for better time control.
In essence, while more trial points can improve the search for an optimum, it's crucial to find a balance that doesn't compromise the effectiveness of time constraints like maxTime.

Products


Release

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!