Brandon T. Willard

> What do the `TransformedX` distributions with `X` being a distribution registered with `_logprob` correspond to? In this case I think we should exclude them. Yes, we definitely need to...

> For distributions we also need a way to specify usage: the links refer to the `Op` documentation, but the way to instantiate these `Op`s (e.g. `at.random.normal`) is nowhere specified....

> I have added a directive to list the invertible transforms, but it would be nice to have it displayed as `autoclass` does. The Sphinx API really is not making...

I'm guessing that our canonicalizations still aren't stable between FAST_COMPILE and FAST_RUN and/or some rewrites that we assume are canonicalizations in AePPL actually aren't. We can always address this by...

Are you talking about creating a numerical integration `Op`? We could definitely do that, but that approach falls well outside of the symbolic stuff we're doing here. Plus, that approach...

It looks like this simply hasn't been implemented.

> Of course: https://aesara.readthedocs.io/en/latest/extending/graph_rewriting.html#detailed-profiling-of-aesara-rewrites. > > Alternatively, you can add a breakpoint [here](https://github.com/aesara-devs/aeppl/blob/d0c009d23df93b43e572b9853138b48933d657f8/aeppl/rewriting.py#L330), `aesara.dprint(fgraph)`, note the index of the `SpecifyShape` node, then inspect the node using `node = fgraph.toposort()[index]` and...

I'm seeing this issue with `asv` at 0.6.3 and `asv_runner` 0.2.1 (not sure if the latter is relevant here). Looking at the `asv.results.Results` class I don't see where `Results` gets...