Mauricio Villegas

You could add `input_width` and `input_height` as parameters of `LitClassification`. And then inspect the signature of `eval(model_name)` and exclude `input_width` and `input_height` from the `kwargs` if that model does not...

> dependency injection would make the config.yaml and training command line more complicated Why exactly do you say this? If dependency injection is used, from command line a class can...

Currently that is not possible. Loading from subconfigs only works at argument level, not for each class in a list. A somewhat close alternative that would work now, although not...

Thank you for reporting! This has been fixed with pull request #458.

Types like `list[]` are supported. The subtype could be simple, like an `int`, or complex like a dataclass or a base class to accept subclasses. See for example pytorch lightning's...

In principle it should be supported. But I have never tried having a subclass type in an `__init__` parameter a class given to `jsonargparse.CLI`. Could be that there is a...

The problem is that the definition `myobj: MyClass` is interpreted as required/positional. When parsing the arguments of `python example.py --config config.yaml subcommand` after parsing the config (optional), it then tries...

With pull request #444 a better error message is given when a class has a `subcommand` method conflicting with the key where the chose subcommand is stored.

Thank you for the proposal! It is an interesting idea. I need to think about it to see if it could have unexpected consequences. One small comment for now. One...

@gautiervarjo thank you for the proposal! I do have in my list of things to improve, a way for the developer to specify which classes to consider or not as...