So Swapnil and I we were doing a stack of different comparisons with others’ software, and it dawned on me, the asymmetic-symmetric variant in Mallet’s LDA approximates HDP-LDA, first described in the hierarchical Dirichlet process article, the much cited non-parametric version of LDA. In fact since 2008 its probably been the best implementation and no-one knew, and its blindingly fast too. Though we got better performance, see our KDD paper.
Bayesian Models
Insights into components of prior research and aspects of latent academia
-
You are currently browsing the Bayesian Models blog archives for June, 2014.
Wray Buntine
Personal Links
Bayesian Models 