Rémi Louf

Now that [Generators are supported in Numba](https://numba.readthedocs.io/en/stable/reference/numpysupported.html#random), we can go ahead and support `RandomStream` in the Numba backend.

I looked into it, and we want to wait at least until the `css` branch is merged on their side: https://github.com/Textualize/textual as many things are about to change.

Yes, if you print `norm_et` defined above you get: ```python norm_et # e(e(, e(, normal, 0, (0, 0), floatX, False)), RandomGeneratorSharedVariable(), TensorConstant{[]}, TensorConstant{11}, TensorConstant{1}, TensorConstant{1}) ``` i.e. `etuplize(at.random.normal)` does wrap...

The following works: ```python from dataclasses import dataclass from aesara.graph import Op, Variable import aesara.tensor as at from aesara.tensor.random.basic import RandomVariable from cons.core import ConsError, _car, _cdr from etuples import...

Finally, just making sure this solves the motivating issue (https://github.com/aesara-devs/aemcmc/pull/29) with running `kanren` goals forward and backward: ```python from kanren import eq, lall, run, var def normaleo(in_expr, out_expr): mu_lv, std_lv...

Some tests that are unrelated to `etuplize`ation expectedly fail. This is easily fixed by wrapping an op with `MakeRandomVariableNodeOp` only if it is a subclass of `RandomVariable`, but still it...

Codecov complains about a file that is not affected by the changes. @brandonwillard this is ready for review.

We *should* be able to unify `etuple(op, ...)` with the result of `etuplize(op(...))`, so I don't think this solution (wrapping the Ops) is acceptable. I gave it more thought, and...

@brandonwillard I made the necessary changes following https://github.com/pythological/etuples/pull/21. Provided the tests pass this is ready on my end.

I considered that, it's just that the apply node created by `RandomVariable`s have the rng as well as the output variable as outputs, and we're only interested in returning the...