Options 'ScaleProblem' in Levenberg-Marquardt-Algorithm (lsqnonlin)

Question

Mazen Slimi 2017-8-31

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/354698-options-scaleproblem-in-levenberg-marquardt-algorithm-lsqnonlin

回答： Veda Upadhye 2017-9-7

I'm using the Algorithm Levenberg-Marquardt (lsqnonlin) in the Optimization Toolbox. And my problem is poorly scaled. By changing the option 'ScaleProblem' from 'none' to 'jacobian', it seems that my function converges better. But what does this option exactly do? And are there more other alternatives to 'none'?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Veda Upadhye 2017-9-7

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/354698-options-scaleproblem-in-levenberg-marquardt-algorithm-lsqnonlin#answer_280623

Hi,

According to the documentation, setting the 'ScaleProblem' option to 'jacobian' sometimes helps the solver on badly-scaled problems i.e it helps improve the convergence of a poorly scaled problem. The other default option is 'none'.

Hope this helps!

Veda