A more detailed Signature property for Ops

[ferrine]

pytensor/pytensor/tensor/random/basic.py

Line 398 in 2ebfbf1

`(), () -> ()`

Will it be useful to make something like this as an Op property? I'm thinking of vectorize rewrites and what they should rely on

[michaelosthege]

If broadcasting needs to be accountable for, this is exactly what infer_shape does, right?

[ricardoV94]

I am not sure I understand the question. This just says how an Op should be vectorized, it doesn't tell what the core shape will be.

This together with infer_shape of the core Op would tell you what the shape of such vectorized Op would be with batched inputs, without having to modify that infer_shape logic.