Nested sampling in practice
Jonathan M. Botts · The Journal of the Acoustical Society of America · 2013
For problems requiring both model comparison and parameter estimation, nested sampling is an attractive choice because it provides an estimate of the normalized posterior. The critical assumption of nested sampling is that exploration proceeds regularly toward the region of maximum likelihood. However, in practice, achieving such regular compression is far from trivial, particularly for realistic, multi-modal problems. This paper offers a comparison of both systematic and random walk exploration for generating samples within a constrained prior. Ranges for size of ensemble and amount of exploration required for regular compression are also established for different acoustic data analysis problems.