LabelTensorizer.ConfigΒΆ

Component: LabelTensorizer

class LabelTensorizer.Config[source]

Bases: Tensorizer.Config

All Attributes (including base classes)

is_input: bool = False
column: str = 'label'
The name of the label column to parse from the data source.
allow_unknown: bool = False
Whether to allow for unknown labels at test/prediction time.
pad_in_vocab: bool = False
Whether vocab should have pad, usually false when label is used as target.
label_vocab: Optional[list[str]] = None
The label values, if known. Will skip initialization step if provided.
label_vocab_file: Optional[str] = None
File with the label values. This can be used when the label space is too large to specify these as a list. The file should not contain a header.
add_labels: Optional[list[str]] = None
Add these labels to the vocabulary during the initialization step (only if the initialization step is not skipped). Useful when the dataset may not include all labels, as for incremental trainings.
Subclasses
  • PositiveLabelTensorizerForDenseRetrieval.Config
  • LabelListRankTensorizer.Config
  • LabelListTensorizer.Config
  • SoftLabelTensorizer.Config

Default JSON

{
    "is_input": false,
    "column": "label",
    "allow_unknown": false,
    "pad_in_vocab": false,
    "label_vocab": null,
    "label_vocab_file": null,
    "add_labels": null
}