Closed twiecki closed 2 years ago
That would be the role of a higher-level library like PyMC imo. @brandonwillard ?
The entry point used by the other Aesara PPL projects involves model graphs and their corresponding value variables, so we should have a similar entry point for samplers—including AeHMC's.
We have code for such an entry point to AeHMC already (including transformations), but still need to decide how to separate/interface it with AeMCMC. It's possible that adding such machinery here would be redundant when/if it exists in greater generality in AeMCMC, so we need to consider that first.
This will most likely happen in aemcmc
, closing for now.
Currently there's a lot of code to set up the kernel etc, would be nice to wrap this into a high-level function similar to
pm.sample()
.