How to pass a type(not an object) to a class with lightningcli?

[calvinzhan]

I want to pass a type(not an object) to a subclass of LightningDataModule. I did the following, but it seemed lightningcli tried to initialize an object and complain required fileds were missing. How can I do that?

class TxtDataModule(LightningDataModule):
    def \__init\__(
        self,
        domain_dataset: Dataset,
        ...
    ):
       dataset_obj = domain_dataset(....)
       ...

in yaml

data:
  class_path: src.training.data_module.data_module_txt.TxtDataModule
  init_args:
    domain_dataset: src.training.semantic_sim.dataset_txt.TxtDataset_LoadFromFile  # a subclass of Dataset

[calvinzhan]

I don't know why I couldn't set label for my discussion.

[tshu-w]

You can change domain_dataset type hint from Dataset to Type:

Here is a demo:

from typing import Type
from jsonargparse import CLI

class Class1:
    def __init__(self) -> None:
        print("Class1")

class Class2:
    def __init__(self) -> None:
        print("Class2")

def test_fn(
    class_type: Type = Class1,
):
    print(class_type)
    class_type()

if __name__ == "__main__":
    CLI(test_fn)

python test.py --class_type '__main__.Class2'

[calvinzhan]

Wow，works again!!!