I have a lot of experience with classical machine learning, and a little bit of experience with Bayesian methods (mostly using pymc3). I mainly want to be able to implement hierarchical classification models, and I’m not clear on how to do this in Edward.
Most of the classification examples rely on using an existing implementation of a dense layer (e.g. denseflipout), which doesn’t give you control over the variables unfortunately.
Would someone be able to write a very short example showing a hierarchical classification model with just a single random-effect please?