Jesse Grabowski

Haha yeah, I stepped through them both in debug mode side-by-side and took notes. That's the only way I got this far. I have a suspicion I'm being a knucklehead...

I believe the following is an implementation using `OpFromGraph`: ```python n = at.iscalar('n') m = at.iscalar('m') k = at.iscalar('k') i = at.switch(k >= 0, k, -k * m) eye =...

How would I check for rewrites that target `Eye`? I did ctrl+f on all the files in `aesara.tensor.rewriting` for `Eye` (and `@node_rewriter([Eye])` ) and came up with nothing. This doesn't...

I'll make a pull request for this in a minute, I'm just fumbling around with git at the moment.

Reporting back on the three suggestions: 1. Passing `Mutable = False` in the third example fixes the bug. 2. Passing a shared variable to size in the first example causes...

As an additional test, I changed the mean of the prior distribution I put over the estimated parameters. It seems that, when using `model.add_coord`, `pm.sample_posterior` actually samples from the prior....

Here are some more through outputs to show that `pm.sample_posterior` is sampling from the prior for all variables, and that the trigger is a shared variable on the shape: ```python...

Having looked at a couple cases where this problem occurs, the main problem is the heavy use of index mappings to "expand" estimated parameters to match data. While it's true...

Well I went and tested it carefully, and @ricardoV94 and @lucianopaz, you're totally right. I will slink off back to my hole of shame. I now agree that this is...

Yeah I'd say so. It's come up enough lately that something should be explicitly stated when you sample the ppc. I think a little communication can go a long way...