Improved configuration parsing

[sarthakpati]

Is your feature request related to a problem? Please describe.
Currently, the configuration parsing is done by a single submodule [ref], which is creates issues with maintenance as more functionality is introduced.

Describe the solution you'd like
Using a configuration validation mechanism that is more standardized would make the developer and user experience much better. Some examples are (and this is by no means an exhaustive list):

Package	Documentation	Example
Cerberus	https://docs.python-cerberus.org/	https://stackoverflow.com/a/46626418
PyDantic	https://docs.pydantic.dev/latest/	https://stackoverflow.com/a/61021183
Marshmallow	https://marshmallow.readthedocs.io/en/stable/	https://stackoverflow.com/a/63739747

Describe alternatives you've considered
N.A.

Additional context
It would make sense to eventually move GaNDLF's various functionalities to something like Hydra, and this could potentially be a good starting point.

[vedik2002]

Hi @sarthakpati can you please assign this issue to me.

[sarthakpati]

Thanks, done.

[github-actions]

Stale issue message

[github-actions]

Stale issue message

[sarthakpati]

Hi @vedik2002, are you still working on this?

[github-actions]

Stale issue message

[sarthakpati]

Hey @vedik2002 are you still working on this?

[vedik2002]

Hi @sarthakpati can please remove me from this issue.

[sarthakpati]

Here are a few viable alternatives to our current configuration management system:

1. Data Classes:

   • dataclasses module [ref] provides a decorator and functions for automatically adding special methods to user-defined classes. This can make your code cleaner and more efficient when dealing with a large number of parameters.

   from dataclasses import dataclass
   
   @dataclass
   class Parameters:
       param1: int
       param2: float
       param3: str

2. Named Tuples:

   • collections.namedtuple module [ref] can be used to create tuple subclasses with named fields. This can be faster than dictionaries for certain operations.

   from collections import namedtuple
   
   Parameters = namedtuple('Parameters', ['param1', 'param2', 'param3'])
   params = Parameters(param1=1, param2=2.0, param3='three')

Did you have something else in mind, @VukW?

[VukW]

I'd prefer to stick to pydantic models (BaseModel, BaseSettings) as they actually give the same experience and simplicity as dataclass but with additional handy tools like custom fields validation, runtime type checking, etc

[sarthakpati]

Do you mean these?

• https://docs.pydantic.dev/latest/api/base_model/
• https://docs.pydantic.dev/latest/api/pydantic_settings/

[VukW]

@sarthakpati yes, exactly

[sarthakpati]

Cool, thanks. 👍🏽