Content deleted Content added
→See also: link |
|||
Line 23:
==Design of Experiments==
The design of computer experiments has considerable differences from [[design of experiments]] for parametric models. Since a Gaussian process prior has an infinite dimensional representation, the concepts of A and D criteria (see [[Optimal design]]) , which focus on reducing the error in the parameters, cannot be used. Replications would also be wasteful in cases when the computer simulation has no error. Criteria that are used to determine a good experimental design include integrated mean squared prediction error [http://projecteuclid.org/DPubS?service=UI&version=1.0&verb=Display&handle=euclid.ss/1177012413] and distance based criteria[http://www.sciencedirect.com/science/article/pii/037837589090122B].
Popular strategies for design include [[latin hypercube sampling]] and [[low discrepancy sequences]].
===Problems with Massive Sample Sizes===
Unlike physical experiments, it is not uncommon for computer experiments to have thousands of different input combinations. Because the standard inference requires [[inversion| matrix inversion]] of a square matrix of the size of the number of samples (<math>n</math>), the cost grows on the <math> \mathcal{O} (n^3) </math>. Currently, this problem this problem is avoided by using approximation methods, e.g. [http://www.stat.wisc.edu/~zhiguang/Multistep_AOS.pdf].
==See also==
*[[Simulation]]
| |||