Thanks for your work on this project, it has been working great for us.
There's a small thing about the handling of the nb_workers argument of pandarallel.initialize I'd like to suggest.

Currently the default value is generated at the top of the file as a global, then set as the default argument of the function. This means if my program has an n_cpus-type Optional argument, I can't just set this as the nb_workers argument as this would overwrite the default value and break things.

Ideally it would be

class pandarallel:
    @classmethod
    def initialize(
        cls,
        shm_size_mb=None,
        nb_workers=None,
        progress_bar=False,
        verbose=2,
        use_memory_fs: Optional[bool] = None,
    ) -> None:

    nb_workers = (
            nb_workers if nb_workers is not None else NB_PHYSICAL_CORES
    )

This would also match the argument handling of use_memory_fs.

I can raise a PR for this if there's interest