Hi there,

I am trying to implement spatial models that use an ICAR prior. There is an issue related to the ICAR prior currently open, and I think the issue is still open because it was hard to test the log probability function because the multivariate distribution itself is a singular Gaussian.

I am trying to use the `pm.Potential`

to test the log probability for various adjacency matrices. I don’t know how to use `pm.Potential`

, but I believe it should be able to be used to replicate a Stan model such as:

from this Stan case study. I have never had a parameter vector within a PyMC model before that I haven’t been able to define by calling a PyMC distribution to specify its prior. It feels like I need to call some Aesara function to define the `phi`

vector as a set of empty? nodes on the graph (but I don’t know what I am talking about ).

If anyone has any thoughts, knowledge, or tips on using `pm.Potential`

within a model, that would be appreciated.

Also, I would love to learn more about Aesara, and if you have any pointers to valuable parts of the Aesara documentation, etc., for this type of thing, that would be awesome.

Thanks,

Conor