Setting a seed for random generators

[nicola-bastianello]

Some users might be interested in setting a seed for random generators (numpy's, pytorch's, etc). Since I think all random generators allow to set a seed, we could extend the kwargs of interoperability random functions to allow for the seed.

[Simpag]

For reproducibility reasons we could remove the seed parameter from the codebase then have a global rng generator for each framework (but only internally used), then we could have one function which can set the seed for all of the global rng generators. Having a singular set_seed reduces the chances for bugs instead of having a seed parameter in multiple places.

This would assure that all randomness in our codebase comes from the same generator and uses the correct seed. Some users might use legacy rng methods, for example in numpy so we could also set those seeds in our set_seed function.

[nicola-bastianello]

Yes, that's a good point, I like the idea of having set_seed like you describe