I am trying to implement KL-divergence in my stochastic block model. When changing the inference ed.MAP to ed.KLqp I get the following error: NotImplementedError: log_prob is not implemented. Only thing I can find on this, is this unsolved issue: https://github.com/blei-lab/edward/issues/803.
I have also tried to change the model parameters to follow a normal dist. instead of PointMass, although this does not work.
Anyone who has tried something similar?