Is there any way to enhance the step of iteration of optimization with fmincon?

3 次查看（过去 30 天）
Arifur Rahman
Arifur Rahman 2018-2-9
回答： Walter Roberson 2018-2-9
Hello. I am trying to do optimization by fmincon. My objection is to minimize the loss. However, when the optimization running, in every iteration the function decrement is so small. Therefore its taking too much time for optimization. Is there any way to enhance the step means decreases with big step? Thanks in advance.
  2 个评论
Walter Roberson
Walter Roberson 2018-2-9
Are analytic gradient and hessian available? If so then passing functions for those in can improve the performance.
Arifur Rahman
Arifur Rahman 2018-2-9
No. The only I have objective value which I have to minimize and there are some lower and upper bounds.

请先登录，再进行评论。

请先登录，再回答此问题。

回答（1 个）

Walter Roberson
Walter Roberson 2018-2-9
According to https://www.mathworks.com/matlabcentral/answers/6694-controlling-stepsize-of-fmincon#answer_266363 you can adjust the FiniteDifferenceStepSize to make it larger, to effectively put a lower bound on the adaptive step size that fmincon uses.
I suspect you will find that it does not make much difference towards obtaining a global minima: what it might do is get it to give up earlier when it encounters a local minima.
If you have the Global Optimization toolbox, you should probably look at MultiStart

类别

在 Help Center 和 File Exchange 中查找有关 Global or Multiple Starting Point Search 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by