Rémi Louf
Rémi Louf
I think (1) can be "solved" when what is being discussed [here](https://github.com/aesara-devs/aesara/discussions/1082) is implemented. Constraints like the one you suggested should be kept out of miniKanren goals imho. Ideally we...
Rebased this on `main` and resolved the merge conflicts that appeared after #1054
Thank you for contributing. I tried to make it run on my machine (with `matrix` and `vector`s instead of shared variables), and the comments are left are the took things...
> N.B. We really should find a way to run `black` on our code snippets in the documentation. And run the examples in the CI! For instance this one is...
Unless I missed something the values of the shared variables were not specified.
> It would probably be easier for the understanding to have everything in the same block. Indeed I always like to copy/paste examples from the documentation when I start playing...
Another option is to let users jit functions themselves. It is more flexible for them, less of a hassle library-side.
> You can transpile without JIT, but that's basically numpy? *No* it's not since you have the possibility to jit-compile the resulting function, and thus specify the `static_argnums` yourself. Many...
From what I've seen in the issue tracker this can be broken down into two separate problems: 1. Shapes whose value is known at compile time by Aesara. In this...
> > Anything else we want to salvage ? > > Looks good to me. Just to be clear, do you mean that we only keep `Softmax`, `LogSoftmax` (which should...