humancompatible.detect.data_handler.features.utils module

humancompatible.detect.data_handler.features.utils.make_feature(data: ndarray | Series, feat_name: str | None, categ_vals: list[int | str] | None, real_bounds: list[int | str] | None, ordered: bool, discrete: bool, monotone: bool, modifiable: bool) → Feature[source]

Factory that creates the appropriate Feature subclass for a single column.

The returned class depends on whether the feature is categorical, binary, mixed, or continuous, and whether a predefined list of allowed categorical values is given.

Parameters:

  • data – One feature column (NumPy array or pandas Series).

  • feat_name – Feature name (may be None).

  • categ_vals – Allowed categorical values. None means continuous.

  • real_bounds – Optional bounds for continuous/mixed features.

  • ordered – Whether categorical values should be treated as ordered (if applicable).

  • discrete – Whether a continuous feature should be treated as discrete.

  • monotone – Whether monotonicity constraints apply.

  • modifiable – Whether the feature is modifiable.

Returns:

A humancompatible.detect.data_handler.features.Feature instance (one of Binary/Categorical/Contiguous/Mixed).