Enable optional component arguments

[PhilippeMoussalli]

PR that enables defining optional arguments to the component spec. This is needed to enable building reusable components.

For example, the load_from_hub component will need to format images from hf format to bytes and this can be achieved by passing a specific argument (image_column_name). This can be achieved either by:

• Passing None explicitly as an argument value at the ComponentOp level
• Assigning None as the default argument field in the component spec.

Previous attempt was to achieve this with the optional field of the kubeflow spec but this returned many errors and also requires adding in an additional optional field to the fondant spec. The current approach achieves the same outcome with a simpler approach.

The component_spec argument is now also defined as an optional argument. This allows use to define a component using the from_file() method without having to explicitly define that argument. We still return a valid error if from_args() is used and no component_spec argument is provided since it defaults to None link

[RobbeSneyders]

Thanks @PhilippeMoussalli, I think this approach is indeed quite elegant! Left one comment and you'll need to rebase / merge :)

[RobbeSneyders]

Maybe it makes sense converting this into a function.

Suggested change

	"String": lambda value: str(value) if value != "None" else None,
	"Integer": lambda value: int(value) if value != "None" else None,
	"Float": lambda value: float(value) if value != "None" else None,
	"Boolean": lambda value: ast.literal_eval(value) if value != "None" else None,
	"JsonObject": lambda value: json.loads(value) if value != "None" else None,
	"JsonArray": lambda value: json.loads(value) if value != "None" else None,
	kubeflow_to_python_type_dict = {
	"String": str,
	"Integer": int,
	"Float": float,
	"Boolean": ast.literal_eval,
	"JsonObject": json.loads,
	"JsonArray": json.loads,
	}
	def kubeflow_to_python_type(type_: str) -> t.Any:
	map_fn = kubeflow_to_python_type_dict[type_]
	return lambda value: map_fn(value) if value != "None" else None

[GeorgesLorre]

or use the match case syntax: https://peps.python.org/pep-0622/

def kubeflow_to_python_type(type: str):
    match type:
        case "String": lambda value: str(value) if value != "None" else None
        case "Integer": lambda value: int(value) if value != "None" else None
        case "Float": lambda value: float(value) if value != "None" else None
        case "Boolean": lambda value: ast.literal_eval(value) if value != "None" else None
        case "JsonObject": lambda value: json.loads(value) if value != "None" else None
        case "JsonArray": lambda value: json.loads(value) if value != "None" else None
        case _: raise ValueError(f"Unknown type: {type}")

[GeorgesLorre]

ignore the above match case is python 3.10+ and fondant is 3.8+

[PhilippeMoussalli]

Thanks for the feedback :) I think it's ready to go

[GeorgesLorre]

thx!