Hello,
Let’s say I have a plate diagram where A → B and C → B and I want to do VI using this model formulation. What are the best practices for incorporating the A and C priors when defining the B distribution?
I know that amortization is an option, but should that be preferred?
Thanks!
it’s hard to give general tips because the details tend to matter a lot. how high dimensional are A, B and C? how informative are the data about A, B, and C? etc etc
That is fair. In my case its a mix.
For example in one scenario I am facing, A has less than 10 dimensions, while B and C have over 1,000