parallel in win notebook

[mszhanyi]

It is very tricky to run multiprocessing in notebook on Windows.
First, The multiprocessing can only be executed in Top-level script environment, that is in the scope of if __name__ == "__main__":.
Otherwise, notebook will hang due to reimporting modules.

Second, all functions, objects used in the subprocesses must be imported from separate module, not from the main module. Because the main module in newly spawned process in notebook is a empty module. If something used in subprocess but is defined in main module, it will throw AtrributeError: Can't get attribute ....

Thirdly, we have to import libraries in the function. otherwise, we'll get an exception that module isn't defined. https://github.com/mszhanyi/fastcore/blob/e1536a5dddbde477e84dfa57e372e4f481353087/fastcore/parallel.py#L128

Lastly, lamda function couldn't be passed to subprocess.

This PR is to provide a example on how to run fastcore.parallel in notebook on Windows.
As for dataloader, It is much harder because there are multiple functions and objects send to workers. code would become very ugly.

@jph00 @maxluk

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[jph00]

Thanks @mszhanyi - sorry hadn't noticed this wasn't merged yet

[renato145]

I noticed that now parallel fails with partial functions.
parallelable checks f.__module__ which partial functions don't have.

[jph00]

Thanks @renato145 . @mszhanyi is this something you might be able to resolve? I guess if it's partial, or for whatever reason __module__ isn't available, we should just assume it's not Windows.