Is it possible to implement online variational Bayes (VB) algorithm for Latent Dirichlet Allocation (LDA) in edward?
I saw that it’s possible to implement LDA with all the following inference methods:
Mean field variational methods (Blei et al., 2001, 2003)
I Expectation propagation (Minka and Lafferty, 2002)
I Collapsed Gibbs sampling (Griffiths and Steyvers, 2002)
I Distributed sampling (Newman et al., 2008; Ahmed et al., 2012)
I Collapsed variational inference (Teh et al., 2006)
I Stochastic inference (Hoffman et al., 2010, 2013; Mimno et al., 2012)
I Factorization inference (Arora et al., 2012; Anandkumar et al., 2012)
I Amortized inference (Srivastava and Sutton, 2016)
Which one can be used with edward / tfp ?
This may answer at-least some of your questions