Hello, I found a great implementation of PLDA in Mahout. However there were certain things that were not clear to me. Firstly the "state" of the LDA that is spit out by each iteration of the EM algorithm (distributed EM) is just the
global state (p(w|t)) 1) The document representations do not constitute the state (or is not written out by the Mahout package) ? 2) How does inference work for new documents work in Mahout, is the support still not available? If any of you have been using PLDA in Mahout, your inputs will be greatly appreciated. Regards, Shivani -- Research Scholar, School of Electrical and Computer Engineering Purdue University West Lafayette IN web.ics.purdue.edu/~sgrao
