Non-collapsing Maximin Designs for Hyper-Polygonal Regions Danel Draguljić

Many researchers use computer codes as substitutes for physical experimentation when the latter is economically or ethically infeasible. Non-parametric predictors of computer code output are typically fitted to training data generated from the code to allow extensive exploration of the input-output relationship. The accuracy of such flexible predictors is compromised unless the inputs for the training data are as ‘space-filling’ as possible. This paper describes an algorithm for constructing designs that maximize a convex combination of the minimum interpoint and the average reciprocal interpoint distances when the input region is a bounded polytope.