Concern about changing the code

[nikorose87]

Hi there,

Your code has been very practical for me. I want to thank you, at first, for this. Second of all, I have so many problems with high dimensionality. Then I would like to implement the following technique for the Meta-model building:

link to the KPLS

In terms of mathematical, numerical and computational viability, is it possible to change the GP to the KPLS in your code?

[ppgaluzio]

Hello, thank you for using the code, I am happy to know it is being useful.

About your question, I think what could be done is to wrap the GP interface into a class, so that to implement the KPLS all you had to do was to wrap its code in a class with the same interface and replace it in the code.

I think that is the easiest way to implement what you need. Anyway, this is a good idea because I was planning to implement some more efficient GP regressors in the future, and having it wrapped in its own class makes that so much easier. And that approach wouldn't require much refactoring now. I will look into that, and see when I can try to do that.

[ppgaluzio]

Hello,

If you look at the readkernel branch, I implemented a wrapper around the KPLS. It is a very simple implementation but it should work, you would have to set the parameters inside the class, or do appropriate changes to the code if you want to be able to pass parameters in the optimizer call.

[nikorose87]

Hello,

If you look at the readkernel branch, I implemented a wrapper around the KPLS. It is a very simple implementation but it should work, you would have to set the parameters inside the class, or do appropriate changes to the code if you want to be able to pass parameters in the optimizer call.

Oh! nice. I will give a try.

[nikorose87]

Hi there, I am back trying to test that wrapper. So, If I uncomment those lines of the wrapper, the KPLS will work with MOBOPT?