July, 2012

21
Jul 12

Numerically stable sampling of the von Mises Fisher distribution (and other tricks)

The von Mises-Fisher distribution is a popular all-purpose distribution for statistical inference involving directional data. Applications in computer graphics involve fitting and approximate sampling of functions on the unit sphere.

Unfortunately, many basic operations involving this distribution are prone to severe numerical issues when implemented in finite precision computer arithmetic. There is a surprising lack on information on how these can be circumvented. I’ve written a short document with the purpose of serving as a collection of numerically-well behaved recipes for common operations.



Link: Numerically stable sampling of the von Mises Fisher distribution on S2 (and other tricks)