How to use the established meta-model (such as Kriging) as Conventional Function in Matlab?

[GPLai (Zhipeng Lai)]

Hello everyone. I’m new UQ man.
I used UQLab to establish a Kriging metamodel. And I want use this metamodel to find the position where the variation is maximum, by adopting Genetic Algorithm. Then I use this maximum varation point as next sample point, evaluate the interested quantity at this new sample point, and refine my meta-model sequentially.
Could anybody give some advices? Where can I find the cases that similar to my situation?

[damarginal (Damar Wicaksono)]

Hi @GPLai,

Referring to the title of your question, I assume you want to use the predictor part of Kriging. As you know, this prediction is made by calling uq_evalModel on a UQLab model and metamodel objects (not just Kriging).

So the easiest way to directly call this predictor as a MATLAB function is by using an anonymous function to wrap the call to uq_evalModel. Let’s suppose you already have a Kriging metamodel object myKriging, then you can write:

myKrigingPredictor = @(X) uq_evalModel(myKriging,X);

Then you can pass this function handle to your optimization function.

However, if you want to get the Kriging variance, which is the second output, for an optimization purpose, I think you might need to use a MATLAB normal function to wrap things first before passing the function handle to an optimization function. For instance, such a function can be:

function krigingVar = getKrigingVar(KrigingObj,X)
[~,krigingVar] = uq_evalModel(KrigingObj,X)
end

and then use an anonymous function to bring a particular Kriging object (say, myKriging) into the function: krigingVar = @(X) getKrigingVar(myKriging,X) before passing the handle to an optimization function.

BTW, There was once a question similar to yours here:

Model-based sampling for more accurate metamodels?

Hi everyone, I’ve been using UQLab for quite some time now to perform sensitivity analysis on an expensive COMSOL model and to generate predictors for my complex model. But now I switched to a matlab model and I’m trying to evaluate as less points as possible to build an accurate metamodel. I was wondering if one specifies an m-file as the input model to UQLab, can UQLab generates model-based samples that can then be used to build a better metamodel. Because the way I understood it, the functi…

UQLab does not have a built-in support for such a refinement scheme to create a metamodel, although a similar scheme is used specifically in the context of reliability analysis. I think your approach makes sense. In any case, we’ll sure be glad to hear about your experience on whether you manage to improve the metamodel .

Hope this helps!

PS: Your question would be more fitting to be in the UQLab Community Q&A, and not in the general forum section. So I took the liberty of moving it there .

[GPLai (Zhipeng Lai)]

Hi @damarginal . Thanks for your nice reply. Yes you do understand my points. And I have found the solution for it after I asked question here. And my solution is same as you proposed here, and it does work. BTW, UQLab is really a nice tool!