Brandon T. Willard

> Why would we need to recognize that a sum of independent Poissons is a Poisson? That's just one fairly straightforward approach to handling this exact situation, but there are...

You're right, that one identity won't work in general. What I'm talking about is the general approach of operating in a non-(log-)density space, for which that convolution identity is just...

> @brandonwillard @rlouf Please let me know which of the conjugates should be more prioritized. It would probably be best for us to get all the exponential family models out...

> Sorry, have not been on top of this. (After Covid and rounds of strep ), I will try to add at least one conjugate a day from now on....

> More concerning to me is how to handle the rngs in these one-to-many relationship. I currently set the rng to both InverseGammaRV to the Half-Cauchy's, but that can't be...

A directly relevant Discussion has been started here: https://github.com/aesara-devs/aesara/discussions/1082#org8a1a145.

Is the basic idea that we should generate log-probabilities with AePPL and apply the transforms to those graphs using AeMCMC, instead of doing all that together in AePPL?

We've converged on the notion of only implementing the core functionality and `numpy.random.*` implementations in this repository, so we might need to move this to AePPL or actually start working...

> But I also suggested there might be no good reason to implement a specific Op for the Frechet, unless the scipy RNG is doing something fancy that is not...

> Implement a `Frechet` Op, say in AePPL, so that the following graph: > > ```python > x_rv = aeppl.random.Frechet(1, 1, 1) > ``` > > is equivalent to: >...