Brandon T. Willard

> I can see that I am in fact out of my league here then, cheers and good luck! There's a bit of context here that may make things seem...

@wrq, here's a MWE demonstrating the underlying issue: ```python from functools import reduce from itertools import chain from toolz import interleave def bind(z, g): # Both fail for the same...

> Ok. First, I'm sorry that I haven't gotten to this sooner. I'm actually a part-time Walmart worker and a stay-at-home father, not really a professional programmer, so this has...

I think we might need to "flatten" this example and see what can be done from there. That would involve combining the `interleave`, `bind`, and `reduce` into a single generator...

> How would we go about having backend-specific rewrites? Numba mode already specializes its rewrites, so check out its definition in `aesara.compile.mode`.

> _Edit:_ Someone on the JAX team confirmed this is a fundamental limitation of HLO that has no dynamically sized arrays. Yeah, I recall them handling things like this in...

> The JAX backend now has better support for `Scan`. I have the following remarks/questions: > > * `mit_mot`s are not currently supported. Should I add this? MIT-MOTS arise from...

> * Setting `a` as `static_argnum` when jit-compiling; this is not the best solution since the function will be re-compiled every time the value of `a` changes, but this mechanism...

This was covered in some Gitter chats (e.g. some egraph conversations [here](https://matrix.to/#/!QqFchLNQAbmeuhDMrX:gitter.im/$vk5Jg2nem90asaynqx2dJS8lC4vyhD-VxXrBFW6RsJc?via=matrix.org&via=gitter.im)) months back, but it should be here as well: with these changes in place, we can more efficiently—and...

> So from which module should one import for instance the convolution operator once that sub-package is indeed removed? We have discussed moving the `nnet` material to another package, but,...