Hello,
I would like to build a Kriging metamodel of a vector-valued model (say, a map from \mathbb{R}^d to \mathbb{R}^n) by constructing an independent Kriging metamodel for each model component, with common experimental design and trend basis (in this way, a set of covariance model hyperparameters is estimated for each output).
It is obviously possible to do so by using a loop on the n model output components, but then, evaluating the n Kriging metamodels might be very slow by using a similar loop. I was wondering if OpenTURNS enables to build n independent Kriging metamodels (one per model output component) for subsequently gathering them into a vector-valued metamodel, which may be evaluated in a vectorized fashion (and then avoiding to evaluate the n Kriging metamodels through a slow loop) ?
Thanks in advance for your attention,
Best regards,
Fab