use a plain serializer to enable pystac Classificiation in BaseModel

[rbavery]

Tests pass for serializing the pystac item and reading in the item with pystac when using the annotated Classification class from pystac in the output BaseModel. If this approach looks good I can do the same for Statistics and Band @fmigneault

[github-actions]

Hello @rbavery, thank you for submitting a PR!

[fmigneault]

I wonder if a model_serializer decorator would not be more explicit and help users understand what is happening?

Would something like this work?

class MLMClassification(BaseModel, Classification):
    @model_serializer()
    def serialize_model(self) -> Dict[str, Any]:
        return self.to_dict()

[rbavery]

Replacing the Annotated type with this class doesn't work unfortunately

/home/rave/work/dlm-extension/tests/test_schema.py::test_model_metadata_to_dict failed with error: Test failed with exception
tests/test_schema.py:8: in mlmodel_metadata_item
    model_metadata_stac_item = eurosat_resnet()
stac_model/examples.py:107: in eurosat_resnet
    class_objects = [
stac_model/examples.py:108: in <listcomp>
    MLMClassification.create(value=class_map[class_name], description=class_name, name=class_name)
../../.cache/pypoetry/virtualenvs/stac-model-YQXQDXJF-py3.10/lib/python3.10/site-packages/pystac/extensions/classification.py:112: in create
    c = cls({})
E   TypeError: BaseModel.__init__() takes 1 positional argument but 2 were given

I suggest for now we stick with this since it works, we can always change the underlying implementation in the future if it is confusing for contributors.

[fmigneault]

If we defined class MLMClassification(BaseModel, Classification) (see other comment), would it be possible to use MLMClassification(value=...) and MLMClassification.create(value=...) interchangeably? Maybe a __init__ calling create could be required since the fields would not be defined.

[rbavery]

I think the use of Annotated is clear enough that we can stick with it. We could override the init, but I think the use of Annotated is more succinct.

[fmigneault]

I have tried using the Annotated approach, but I ended up getting multiple unhashable dict errors when using MLMClassification within another BaseModel definition.

[fmigneault]

I had to do the following to make it all work together:

dlm-extension/stac_model/output.py

Lines 34 to 65 in 2d6c70b

	class MLMClassification(BaseModel, Classification):
	@model_serializer()
	def model_dump(self, *_, **__) -> Dict[str, Any]:
	return self.to_dict()
	def __init__(
	self,
	value: int,
	description: Optional[str] = None,
	name: Optional[str] = None,
	color_hint: Optional[str] = None
	) -> None:
	Classification.__init__(self, {})
	if not name and not description:
	raise ValueError("Class name or description is required!")
	self.apply(
	value=value,
	name=name or description,
	description=description or name,
	color_hint=color_hint,
	)
	def __hash__(self) -> int:
	return sum(map(hash, self.to_dict().items()))
	def __setattr__(self, key: str, value: Any) -> None:
	if key == "properties":
	Classification.__setattr__(self, key, value)
	else:
	BaseModel.__setattr__(self, key, value)
	model_config = ConfigDict(arbitrary_types_allowed=True)

[rbavery]

We don't need to include MLMClassification in any other BaseModel definitions in this code base though I think? So it shouldn't be a problem? I haven't encountered that error when using Annotated when serializing or loading the data back in with pystac. The class above is more verbose and complex for contributors than the use of Annotated imo. Can we keep it as is since it works?

[fmigneault]

Yes. It must be included in the model output. We employ the same definition as classification:classes, but they are not currently validated by that extension, because the field is nested under a specific MLM output rather than at the Item properties or Asset level (see stac-extensions/classification#48).

Using the above definition, serialize/load works both ways, as evaluated by this test:

dlm-extension/tests/test_schema.py

Lines 36 to 39 in 4eb30da

	def test_validate_model_against_schema(eurosat_resnet, mlm_validator):
	mlm_item = pystac.read_dict(eurosat_resnet.item.to_dict())
	validated = pystac.validation.validate(mlm_item, validator=mlm_validator)
	assert SCHEMA_URI in validated

using this definition:

dlm-extension/stac_model/examples.py

Lines 106 to 116 in 4eb30da

	class_objects = [
	MLMClassification(value=class_value, name=class_name)
	for class_name, class_value in class_map.items()
	]
	output = ModelOutput(
	name="classification",
	tasks={"classification"},
	classes=class_objects,
	result=result_array,
	post_processing_function=None,
	)

[rbavery]

Ok gotcha. I'm still a bit confused why the use of Annotated wasn't working out but trust this is best.

Since you have this working in your PR I'll close this @fmigneault