#  Dirichlet Simplex Exploration: Or, My Prayer Answered 

 





September 20, 2006

 

 

Andrew Fernandes, a fellow Canadian expat and PhD student at NC State, responded to my earlier request for [advice on exploring a Dirichlet-type simplex](http://www.iq.harvard.edu/blog/sss/archives/2006/05/).

Among other places, the idea is presented in the Wikipedia entry for [Simplex](http://en.wikipedia.org/wiki/Simplex#Random_Sampling). He suggests perturbing the cumulative sums, then putting the perturbed sums back in order to draw a time-reversible proposal. This has the advantage of not sending too many parameters below zero - a maximum of one - as opposed to an equal perturbation of each parameter, and not pinning a high-valued parameter in place with a standard Dirichlet proposal.

Posted by [Andrew C. Thomas](http://www.iq.harvard.edu/blog/sss/archives/author/andrew-c-thomas/) at September 19, 2006 11:32 PM



 

 

 



 

 

 Share on:- [     Facebook ](#)
- [     Twitter ](#)
- [     Linkedin ](#)