What software for optimization surface of 4 parameters?

[Tosh5457]

Hello,

I have data for optimization that contains different values of 4 different parameters, and the respective function value based on them. I want to find the maximum of this function, while making sure it's a "stable" maximum, i.e., for nearby parameters, the value of the function shouldn't change that much. My only question is what tool to use for this, since I have 60k values in total? I have experience in Matlab, can that be done in a reasonable amount of time there? I also have some basic knowledge of R, and can learn fast anyway, is this more suitable to do in R?

Thanks!

 

[h6ss]

Hello,

I have data for optimization that contains different values of 4 different parameters, and the respective function value based on them. I want to find the maximum of this function, while making sure it's a "stable" maximum, i.e., for nearby parameters, the value of the function shouldn't change that much. My only question is what tool to use for this, since I have 60k values in total? I have experience in Matlab, can that be done in a reasonable amount of time there? I also have some basic knowledge of R, and can learn fast anyway, is this more suitable to do in R?

Thanks!

I would definitely go with R, even (or I should say especially) with 60,000 values. Also, but depending on the context, I think the Nelder-Mead optimization method would be an interesting approach to try.

I've used the neldermead package before and it served me well.

More information here : https://cran.r-project.org/web/packages/neldermead/neldermead.pdf

 

[Tosh5457]

I would definitely go with R, even (or I should say especially) with 60,000 values. Also, but depending on the context, I think the Nelder-Mead optimization method would be an interesting approach to try.

I've used the neldermead package before and it served me well.

More information here : https://cran.r-project.org/web/packages/neldermead/neldermead.pdf

Thank you, will definitely look into that method in R.

 

[Tosh5457]

Ok, Nelder-mead won't work, because I don't have the function explicitly...

 

[h6ss]

Ok, Nelder-mead won't work, because I don't have the function explicitly...

The explicit form of the function is not necessary. The Nelder-Mead algorithm only requires function values.

The algorithm is often called the "simplex" method, maybe you can find more information about it under that label. Also, many books cover the subject, e.g. Richard Brent's book Algorithms for Minimization Without Derivatives.