Skip to content

serializers

allennlp.evaluation.serializers.serializers

[SOURCE]

Serializer

class Serializer(Registrable)

General serializer class for turning batches into human readable data

__call__

class Serializer(Registrable):
 | ...
 | def __call__(
 |     self,
 |     batch: Dict[str, TensorField],
 |     output_dict: Dict,
 |     data_loader: DataLoader,
 |     output_postprocess_function: Optional[Callable] = None
 | ) -> str

Postprocess a batch.

Parameters

  • batch : Dict[str, TensorField]
    The batch that was passed to the model's forward function.

  • output_dict : Dict
    The output of the model's forward function on the batch

  • data_loader : DataLoader
    The dataloader to be used.

  • output_postprocess_function : Callable, optional (default = None)
    If you have a function to preprocess only the outputs ( i.e. model.make_human_readable), use this parameter to have it called on the output dict.

Returns

  • postprocessed : str
    The postprocessed batches as strings

default_implementation

class Serializer(Registrable):
 | ...
 | default_implementation = "simple"

SimpleSerializer

@Serializer.register("simple")
class SimpleSerializer(Serializer)

Very simple serializer. Only sanitizes the batches and outputs. Will use a passed serializer function for the outputs if it exists.

__call__

class SimpleSerializer(Serializer):
 | ...
 | def __call__(
 |     self,
 |     batch: Dict[str, TensorField],
 |     output_dict: Dict,
 |     data_loader: DataLoader,
 |     output_postprocess_function: Optional[Callable] = None
 | )

Serializer a batch.

Parameters

  • batch : Dict[str, TensorField]
    The batch that was passed to the model's forward function.

  • output_dict : Dict
    The output of the model's forward function on the batch

  • data_loader : DataLoader
    The dataloader to be used.

  • output_postprocess_function : Callable, optional (default = None)
    If you have a function to preprocess only the outputs ( i.e. model.make_human_readable), use this parameter to have it called on the output dict.

Returns

  • serialized : str
    The serialized batches as strings