ner.utils.visualize module#

ner.utils.visualize._get_pred_tags_and_unk_tokens_from_text(tokenizer: Tokenizer, model: Module, text: List, device: device = device(type='cpu')) → Tuple[List[str], ndarray]#

Get the tag predictions (using a model) and unknown tokens from the input list of tokens.

Parameters:

tokenizerTokenizer: Tokenizer.
modelModule: Model.
textList: Text.
devicetorch.device: Device.

Returns:

List[str]: Predicted tags.
np.ndarray: Unknown tokens.

ner.utils.visualize._print_convention()#: Print the color convention used in activation visualization plots.

ner.utils.visualize.inspect_preds(tokenizer: Tokenizer, model: Module, text: List, labels: List | None = None, device: device = device(type='cpu')) → List[str]#

Inspect predictions at a token-level.

Parameters:

tokenizerTokenizer: Tokenizer.
modelModule: Model.
textList: Text.
labelsOptional[List]: Labels.
devicetorch.device: Device.

Returns:

List[str]: Predicted tags.

ner.utils.visualize.visualize_activations(tokenizer: Tokenizer, model: Module, module: Module | Module, text: List, prev_layer_module: Module | Module | None = None, labels: List | None = None, nonlinearity: Callable | None = None, device: device = device(type='cpu'), cbar: bool | None = True, figsize: Tuple[int, int] | None = None, fontsize: int = 8)#

A function to visualize activations by attaching forward hooks to the model.

Parameters:

tokenizerTokenizer: Tokenizer.
modelModule: Model.
moduleUnion[Module, nn.Module]: Module.
textList: Text.
prev_layer_moduleOptional[Union[Module, nn.Module]]: Previous layer module.
labelsOptional[List]: Labels.
nonlinearityOptional[Callable]: Nonlinearity.
devicetorch.device: Device.
cbarOptional[bool]: Whether to show colorbar.
figsizeTuple[int, int]: Figure size: (width, height).
fontsizeint: Font size.