convergence / window size / goodness of fit measures / inference schemes

jwijffels commented 5 years ago

Hi @xiaohuiyan I've created an R package wrapping the C++ at https://github.com/bnosac/BTM which works fine. While doing that, I had a number of questions on the C++

the window size is set to 15. Is there any reason why this is fixed in the C++ code?
can you provide ways or advice on how to see if the gibbs sampler has converged (what about burnin e.g., trace plots, ...)
can we get from the C++ code somehow measure of statistics as provided in the paper?
can you indicate with respect to the paper what the inference schemes sum_b, sub_w and mix refer to
if you do inference on new text, do you need to remove the terms from pw_z which are not part of the new text (I'm confused by this question: https://github.com/xiaohuiyan/BTM/issues/1)

many thanks!

madegomez commented 3 years ago

Hi. Did you find an answer to your last question? The question about inference on new text.

Thanks

jwijffels commented 3 years ago

No

xiaohuiyan / BTM

convergence / window size / goodness of fit measures / inference schemes #24