Brandon T. Willard

> So will we match `at.inv(weibull)` directly in our rewrites? And are we ok not having a `FrechetRV` object we can directly manipulate and represent in the graph? Then, what...

Before I forget, [here's](https://gist.github.com/brandonwillard/10f8bdfe1c307c590472f73500f6ae67) a Gist with the beginnings of an implementation for all this. It has the added constraint that it doesn't want to change the multiplication order of...

Are you guys essentially talking about how the endpoints of the `clip`ed intervals are/should be handled/understood?

> There's an implicit assumption about the meaning of the bounds when using `at.clip` that is not clearly stated, and not completely obvious. Saying that the value is equal to...

This should be covered by `aesara.tensor.random.opt.local_dimshuffle_rv_lift` in a lot of cases, since it would lift the `DimShuffle` to the arguments of the `RandomVariable`. `local_dimshuffle_rv_lift` may not currently support that distribution,...

> You could however extend the lift if the Dimshuffle only affects non-core dimensions Yes, the current version of `local_dimshuffle_rv_lift` even takes this into consideration, which is why the multivariate...

Now that I think about it, there may be a way to make it _always_ possible to lift these operations. The reason we can't lift the `DimShuffle` through `DirichletRV` is...

> Why not just DimShuffle the value variables and evaluate it at the original untransposed variable? While that might work for the purpose of constructing a log-probability function—in the same...

> I don't see why this would be the case with our current eager IR. A dimshuffle of a measurable component would be converted into a MeasurableDimShuffle eventually (since such...

> But the following does not: > > ```python > import aeppl > import aesara.tensor as at > > srng = at.random.RandomStream(0) > > a_rv = srng.uniform(0, 1.0, name="A") >...