reject.uncertainty

Module for uncertainty.

Module Contents

Functions

compute_uncertainty(→ Union[numpy.typing.NDArray, ...)	Calculate total uncertainty (TU), aleatoric uncertainty (AU) and epistemic uncertainty (EU).
compute_confidence(→ numpy.typing.NDArray)	Compute confidence.

reject.uncertainty.compute_uncertainty(y_pred: numpy.typing.NDArray, unc_type: reject.constant.EntropyUnc | None = None) → numpy.typing.NDArray | dict[str, numpy.typing.NDArray]

Calculate total uncertainty (TU), aleatoric uncertainty (AU) and epistemic uncertainty (EU).

Parameters:

• y_pred (NDArray) – Array of predictions. Shape (n_observations, n_classes) or (n_observations, n_samples, n_classes).

• unc_type (Unc_type, optional) – Type of uncertainty to compute (either TU, AU, or EU), by default None

Returns:

Array of one uncertainty type, or all three uncertainty types.

Return type:

Union[NDArray, tuple[NDArray, NDArray, NDArray]]

Raises:

ValueError – If unc_type is invalid.

reject.uncertainty.compute_confidence(y_pred: numpy.typing.NDArray) → numpy.typing.NDArray

Compute confidence.

Parameters:

y_pred (NDArray) – Array of predictions. Shape (n_observations, n_classes) or (n_observations, n_samples, n_classes).

Returns:

conf – Array of confidence values.

Return type:

NDArray