Asynchronous Dataloader: RecursionError: maximum recursion depth exceeded while calling a Python object

[FarzanT]

Hello,
There seems to be a bug where if there's an array of strings, the load_instance method is called again on the sample:

https://github.com/Lightning-Universe/lightning-bolts/blob/0290ab0ad7824a9917935af0975244f01a6e1ca7/src/pl_bolts/datamodules/async_dataloader.py#LL79C3-L81C10

elif elem_type.__module__ == "numpy" and elem_type.__name__ != "str_" and elem_type.__name__ != "string_":
    if elem_type.__name__ == "ndarray" and self.np_str_obj_array_pattern.search(sample.dtype.str) is not None:
        return self.load_instance(sample)

I think the correct version would just return the sample, like so:

elif elem_type.__module__ == "numpy" and elem_type.__name__ != "str_" and elem_type.__name__ != "string_":
    if elem_type.__name__ == "ndarray" and self.np_str_obj_array_pattern.search(sample.dtype.str) is not None:
        return sample

This works in my case. If it's acceptable, please let me know so I'll open a pull request.

[Borda]

@FarzanT, nice finding, would you be interested in seeing fix? 🐰

[FarzanT]

@Borda Of course, I’m on vacation, but I’ll slowly start a pull request. It’s a single line change.