Some Computational Details of Configural Sampling Methods

摘要

Numerical evaluation of the optimum estimate via configural sampling involves evaluation of several double integrals. These integrals represent expectations over a distribution conditioned on the observed configuration. Theoretically, any location-scale invariant definition of the configuration will suffice, though numerically, some choices are better than others. A related concern is the change-of-variables used to map the region of integration, originally the half-plane, onto a fixed region, such as the unit square. This report is of use to the reader both as a guide to the pitfalls and curiosities of the computations presently recommended. (Author)