hankcs/HanLP · error · NotImplementedError

Not supported.

Error message

Not supported.

What it means

FastTextDataset.load_file is explicitly NotImplementedError because FastText embeddings in HanLP are loaded from a pretrained .bin/vector file via the FastTextEmbedding module, not from an arbitrary corpus file. The dataset only exists to transform in-memory token lists handed to it by build_dataloader; calling load_file directly (or pointing the component at a file path to read as samples) is unsupported.

Source

Thrown at hanlp/layers/embeddings/fast_text.py:113

            src: Field name.
            filepath: Filepath to pretrained fastText embeddings.
        """
        super().__init__()
        self.src = src
        self.filepath = filepath
        self._fasttext = FastTextTransform(self.filepath, self.src)

    def transform(self, **kwargs) -> Optional[Callable]:
        return self._fasttext

    def module(self, **kwargs) -> Optional[nn.Module]:
        return FastTextEmbeddingModule(self._fasttext.dst, self._fasttext.output_dim)


class FastTextDataset(TransformableDataset):

    def load_file(self, filepath: str):
        raise NotImplementedError('Not supported.')


class FastTextEmbeddingComponent(TorchComponent):
    def __init__(self, **kwargs) -> None:
        """ Toy example of Word2VecEmbedding. It simply returns the embedding of a given word

        Args:
            **kwargs:
        """
        super().__init__(**kwargs)

    def build_dataloader(self, data, shuffle=False, device=None, logger: logging.Logger = None,
                         **kwargs) -> DataLoader:
        embed: FastTextEmbedding = self.config.embed
        dataset = FastTextDataset([{'token': data}], transform=embed.transform())
        return PadSequenceDataLoader(dataset, device=device)

    def build_optimizer(self, **kwargs):

View on GitHub (pinned to ddb1299bdd)

Solutions

1. Don't use FastTextDataset to read files; construct it with in-memory data like FastTextDataset([{'token': data}], transform=...)
2. Load fastText vectors via the FastTextEmbedding config instead
3. Subclass TransformableDataset and implement load_file if you need file loading

Example fix

# before
ds = FastTextDataset(['train.txt'])  # triggers load_file -> error
# after
ds = FastTextDataset([{'token': tokens}], transform=ft_embed.transform())

assert isinstance(data, list) and not isinstance(data, str), 'pass in-memory token lists, not file paths'

def is_inmemory_data(data) -> bool:
    return isinstance(data, (list, tuple)) and not any(isinstance(d, str) and d.endswith('.txt') for d in data)

When it happens

Trigger: Instantiating FastTextDataset and calling .load_file(path), or configuring a pipeline that tries to load training samples from a file into this dataset class.

Common situations: Trying to train fastText embeddings inside HanLP; reusing the dataset class as a generic token dataset and passing a filename.

Understand the failure class

Background: UnsupportedOperationException and "is not supported" errors: when a library deliberately refuses a call — this error's family across 30 libraries.

Related errors

• Not supported.
• error
• output ({}) must be of type bool or str
• Call fit or load before evaluate.
• Unrecognized devices {devices}

AI-assisted analysis of hankcs/HanLP@ddb1299bdd (2026-08-27). Data as JSON: /api/errors/d674757e4c1f43ee. Report an issue: GitHub.