On Nov 2, 2011, at 5:31 PM, Ted Dunning wrote: > > For some kinds of models, notably all of the ones from the > exponential class, there exist sufficient statistics and the combination of > models really is a lot like addition. Most of the uses of DP clustering > involve exponential models like the normal distribution so the world is a > happy place.
Can you elaborate on this a bit more in terms of concrete steps we could take to implement this?