Hi, team. I got a flyte workflow that returns a NamedTuple

return ModelOutput(
        model_version=model_version,
        bst_rounds=training_output.bst_rounds,
        classifier=training_output.classifier
    )

And the it is defined as

ModelOutput = NamedTuple(
    "ModelOutput",
    model_version=ModelVersion,
    bst_rounds=int,
    classifier=Classifier
)

It got error when packaging

Transformer for type <class 'tuple'> is restricted currently

However, it used to work when I set the ModelOutput as below, my changes is just add a new filed to the existing tuple. Anyone knows this issue? Thanks in advance

ModelOutput = NamedTuple(
    "ModelOutput",
    model_version=ModelVersion,
    bst_rounds=int
)

FYI, there is another structure in our code that gets to work

TrainingOutput = NamedTuple(
    "TrainingOutput",
    classifier=Classifier,
    model_pickle=FlyteFile[TypeVar("pkl")],
    model_bst=FlyteFile[TypeVar("bst")],
    feature_importance=PNGImageFile,
    bst_rounds=int
)

This is weird. Could you share your code snippet? Will try reproducing the error.

I am also facing similar issue in our project :

"Exception when executing Tuples are not a supported type for individual values in Flyte - got a tuple

hi, @Samhita Alla here it is

@workflow
def train_register_model(
        config: Config,
        all_data: DataSlicer,
        stage: str,
        last_best_rounds: int,
        validate_on_service_type: bool) -> ModelOutput:
    """This workflow trains an xgb classifier and create a new model
    version on Merlin.

    Creation of a Merlin model version implies the registration of the
    model and log the artifacts/metadata on Mlflow.
    """
    training_output: TrainingOutput = train_classifier(
        config=config, all_data=all_data, stage=stage, last_best_rounds=last_best_rounds
    )
    offline_metrics: Dict = compute_offline_metric(
        model=training_output.classifier,
        all_data=all_data,
        stage=stage,
        config=config,
        on_service_type=validate_on_service_type,
    )
    metadata: Metadata = create_metadata(
        config=config, all_data=all_data, metrics=offline_metrics, stage=stage
    )
    model_version: ModelVersion = create_pyfunc_model_version(
        config=config,
        model_pickle=training_output.model_pickle,
        model_bst=training_output.model_bst,
        feature_importance=training_output.feature_importance,
        metadata=metadata,
    )
    return ModelOutput(
        model_version=model_version,
        bst_rounds=training_output.bst_rounds,
        classifier=training_output.classifier
    )

ModelOutput = NamedTuple(
    "ModelOutput",
    model_version=ModelVersion,
    bst_rounds=int,
    classifier=Classifier
)

@Zhang Yue, what's

classifier

here? What's its type?

@Samhita Alla thanks for the reply, i tried with NamedTuple once but will try again

Okay. Let me know if you come across any error with NamedTuple.

hi, @Samhita Alla classifier is a class

class Classifier(ABC):
    @abstractmethod
    def fit(self, *params) -> None:
        pass

wf_outputs = typing.NamedTuple("OP2", greet1=str, greet2=str, classifier=Classifier)

works for me. What's

ModelVersion

?

you mean it is able to package is it?

I ran the

pyflyte run

command and yes, it's working for me. What flytekit version have you installed?