Are there any packages there any LDA packages out there that allow you to specify the full asymmetric prior? I know Mallet uses asymmetric priors, but I don't believe it allows you to specify them.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
The gensim topic modelling toolbox in python seems to support asymmetric priors. This documentation says:
alpha can be set to an explicit array = prior of your choice. It also support special values of ‘asymmetric’ and ‘auto’: the former uses a fixed normalized asymmetric 1.0/topicno prior, the latter learns an asymmetric prior directly from your data.
Hope that helps!
-
$\begingroup$ Definitely helpful! I'd prefer a package that uses Gibbs sampling, but this might be my best bet. $\endgroup$– elplattCommented Mar 18, 2017 at 21:47
-
$\begingroup$ I know that the Matlab topic modelling toolbox uses Gibbs sampling, but does not support asymmetric priors readily. It should not be too difficult to modify their original code to use asymmetric priors. $\endgroup$– kedarpsCommented Mar 18, 2017 at 21:55