GlobalSearch does not keep the best solution

Hi All!

I have a cmplex, non-convex problem with ~30 parameters. I have been using GlobalSearch for a long while now, but since my model got more complex, I have been expereincing this issue, where I can see that GlobalSearch finds some good solutions (let's say they result in ~15% error), but when it keeps running, it doesn't necessarily return those as the local minima. Instead it return solutions with worst results such as 25% error. Any ideas what might be wrong?

Answer 1

[Shashank Gupta]

Hi ghazal,

The optimization becomes very difficult with increase in the dimension or parameter. I am not sure what exactly can be the reason of failure of global search since I don't the what optimization problem you are tackling. Although there are some resources available for you to refer. Check out this link, it will give you the reference on what solver we should prefer depending on optimization problem. Also there is one documentation available to tell us whether we reach the optimal solution or not, check out this link for the same.

I hope this helps.

Cheers.

[Walter Roberson]

There are some situations where fmincon might not return the lowest value for some algorithms. I have not checked to see if the bug/limitation was fixed yet.