Brandon T. Willard

Yeah, we definitely need to do this, because it looks like it will produce instruction messages for issue creators/submitters that will not be included in the resulting issues.

We need to determine if `TopologicalSorter` could be worthwhile to adapt to this use-case.

> ...if python >= 3.10 can be assumed. I believe we're already requiring `typing_extensions`, so `ParamSpec` should be available.

Thanks for the contribution! I'm pretty busy right now, so I might not get a chance to review this in the next few days, but the changes sound great, so...

Profiling graphs with `Scan`s (or any other `Op`s that hold Aesara graphs inside them) is currently complicated by the way in which their inner-graphs are compiled. Simply put, the inner-graphs...

I've used something like the following in the past to read profiling info on `Scan`s: ```python scan_fn = aesara.function(...) for node in scan_fn.maker.fgraph.apply_nodes: if isinstance(node.op, Scan) and hasattr(node.op.fn, "profile"): node.op.fn.profile.summary()...

Thank for the bug report! We'll take a closer look at it shortly.

I have a fix in #93 for the error you were seeing, but we still need to look into the samplers AeMCMC produces (or lack thereof) for your model.

That `CumOp` warning is something we still need to address. We can open another issue for it, though.

> Looked like an error to me? Yes, it is. I'll reopen this and take a look soon.