Add shuffle RandomVariable

[rlouf]

https://numpy.org/doc/stable/reference/random/generated/numpy.random.shuffle.html

[PaulScemama]

What do you have in mind for this? I apologize but I'm not sure what this would entail at a lower-level. For example, call numpy's shuffle? Create our own shuffle with the Fisher-Yates algorithm? Thanks!

[brandonwillard]

What do you have in mind for this? I apologize but I'm not sure what this would entail at a lower-level. For example, call numpy's shuffle? Create our own shuffle with the Fisher-Yates algorithm? Thanks!

When it comes to implementing a new Op, the approach is generally as follows:

1. Determine how much, if any, of the Op can be implemented using existing Ops in Aesara.
2. For the part(s) that can't be implemented using existing Ops, we need to create new Ops, which consists of the following steps:
   • Determine the input and output Types (i.e. the Op.make_node implementation).
   • Determine how the symbolic inputs and their shapes relate to the symbolic shapes of the outputs (i.e. the Op.infer_shape implementation).
   • Determine the symbolic gradients for each input/output combination (i.e. the Op.L_op/Op.grad implementation).
   • Determine the implementations in Python (i.e. Op.perform), C (i.e. Op.c_code), Numba, and/or JAX. Only the Python implementation is absolutely necessary, but the others are needed for performance and the general usefulness of the Op.

[PaulScemama]

Thanks for this overview! I've been a bit busy the last couple days so haven't had a chance to dive in at all, but looking forward to taking a better look this weekend (if not this week).