The result model varies for each run
Opened this issue · 1 comments
GoogleCodeExporter commented
Hi,
I have little knowledge on DP. However, I used the code to estimate the density of some mixtures and try to get the mixture components, It seems that the algorithm produces different results each time. The self.v values differ each time.
I am not sure whether this is a bug or it is the nature of the variational inference algorithm.
Original issue reported on code.google.com by caofan....@gmail.com
on 17 Sep 2013 at 11:23
GoogleCodeExporter commented
I am a little late - Google never notified me of this! Only noticed because it
appeared when I transferred this project to github - sorry.
Anyway, variational methods are initialised with random numbers, and then
converge to a local minima - different initialisations will lead to different
results. So yes, getting different values each time is expected. You still
expect to get something similar however - if there is a massive difference in
the meaning of the output then there is problem, though note that mixture
components can come out in an arbitrary order.
Original comment by thai...@gmail.com
on 26 Mar 2015 at 10:51
- Removed labels: Type-Defect