|
Is there a way to sample beta given the set of pi's in the HDP? In http://www.gatsby.ucl.ac.uk/~ywteh/research/npbayes/hdp2004.pdf, section 5.2, Teh et. al. give a method for sampling beta given a set of auxiliary dish counts. I am interested in knowing if sampling is possible in the fully uncollapsed HDP. It seems like the normalization constraint on beta makes it hard to use a technique like ARS to sample each beta_k individually. |
