Thanks for the talk. That's maybe not the place to ask questions but it will help your referring: Is this pymc_experimental.marginalize only for discrete variables (which were already very easy to marginalise by hand) or is it able to work with some continuous functions? Automatic marginalisation of continuous variables would be awsome! Which is certainly related to the posterior conjugacy you're discussing in conclusion: those are the cases in which marginalisation is easy.
Sorry for not replying before. We definitely want to expand the functionality to continuous marginalization. The finite discrete cases were just easier as a proof-of-concept because they can always be generated.
Sorry if this is obvious but what module is `pt` (the one with exp and abs in the 1-to-1 and many-to-1 transform examples)?
Found it - it’s `pytensor.tensor`.
Thanks for the talk.
That's maybe not the place to ask questions but it will help your referring:
Is this pymc_experimental.marginalize only for discrete variables (which were already very easy to marginalise by hand) or is it able to work with some continuous functions? Automatic marginalisation of continuous variables would be awsome! Which is certainly related to the posterior conjugacy you're discussing in conclusion: those are the cases in which marginalisation is easy.
Sorry for not replying before. We definitely want to expand the functionality to continuous marginalization. The finite discrete cases were just easier as a proof-of-concept because they can always be generated.