How to bulid PCE model for given data?

UQ with UQLab Community Q&A and How To
, ,
1

Hi everyone,

I am wondering how to build PCE model based on given correlated non-Gaussian data.

For example, I have a data matrix with 30 dimensions and 17520 observations. I am assuming the underlying copula is the Gaussian copula. Then what should I do to build a PCE model based on this data? Should I do the Isoprobabilistic transformation myself? Or will the Uqlab automatically do this for me if I specify my input as following:

iOpts.Marginals = uq_KernelMarginals(solar_power’);
C = corr(solar_power’,‘Type’, ‘Spearman’);
iOpts.Copula.Type = ‘Gaussian’;
iOpts.Copula.RankCorr = C;
Input=uq_createInput(iOpts);

Any help or sample code will be appreciated.

Best regards,
Yx Wu

2

I suggest you start from the example: https://www.uqlab.com/pce-truss-data-set

3

Thank you for your reply. But I think this case focuses on useing existing data to replace experimental design, not using exsiting data to define an input object.

For the given input data, I am wondering how will uqlab deal with it inside this toolbox? In my case above, I used kernel estimation to infer the input distribution and I assumed the dependence structure is a Gaussian Copula. Then if I continued with uqlab to build model and PCE metamodel, what will uqlab do? Will it do an isoprobabilistic transformation automatically based on which type of polynomial basis I have assigned?

5

As far as I know, PCE automatically applies transformations. However, I have noticed that they have made many changes and updates. Therefore, I would suggest setting a breakpoint in the corresponding .m file that you are concerned about to double-check. If you are wondering how UQLab deals with it inside its toolbox, the best way to find out is to go through the code along with the manual.

6

Got it! Thank you very much.