templates.MultiPolarModelBasedTemplateSchema

class templates.MultiPolarModelBasedTemplateSchema(*args, **kwargs)

Multipolar Model Based template

Parameters:

• include_fields (List[str]) – The fields to include in the model featurization.

• target_field (str) – The field containing the model predictions.

• model_type (str) –

  The specific model used, must be contained in:

  [“zsl_text_match”, “zsl_entailment”, “masked_lm”, “logistic_regression”, “one_class_svm”, “tfidf_logreg”, “setfit”, “sdnet”, “sequence_embedding”]

• model_name (str) – A user specified string name given to the trained model.

• value (None) – Inherited from ModelBasedTemplateSchema, but not used.

• dirpath (str) – The path to the stored model in minio.

• threshold (float) – The confidence threshold below which this LF should abstain.

• unique_model_name (Optional[str] = None) – A unique name for the model, usually the same as dirpath.

• is_multilabel (bool = False) – Whether or not the task this LF is being applied to is multi-label

• multilabel_absent_as_abstain (bool = True) – If this is a multi-label task, True if we should replace “absent” votes output by the model with “abstain” votes

• inv_label_map (Dict[str, int]) – A mapping from the user’s class labels to the LF’s class labels.