nessie.detectors.baselines

Module Contents

Classes

MajorityLabelBaseline	The majority baseline computes the most common label seen and then simply
MajorityLabelPerSurfaceFormBaseline	This majority baseline computes the most common label seen per surface form and then simply

class nessie.detectors.baselines.MajorityLabelBaseline

Bases: nessie.detectors.error_detector.Detector

The majority baseline computes the most common label seen and then simply flags all items that are disagreeing with it.

correct(self, texts: nessie.types.StringArray, labels: nessie.types.StringArray, **kwargs) → numpy.typing.NDArray[str]

error_detector_kind(self) → nessie.detectors.error_detector.DetectorKind

score(self, texts: nessie.types.StringArray, labels: nessie.types.StringArray, **kwargs) → numpy.typing.NDArray[bool]

supports_correction(self) → bool

class nessie.detectors.baselines.MajorityLabelPerSurfaceFormBaseline

Bases: nessie.detectors.error_detector.Detector

This majority baseline computes the most common label seen per surface form and then simply flags items that are disagreeing with it. This is more useful for token and span labeling, as there are more repeated surface forms. For instance, if Obama has been seen twice as person and once as a location, then the instance with location is getting flagged.

correct(self, texts: nessie.types.StringArray, labels: nessie.types.StringArray, **kwargs) → numpy.typing.NDArray[str]

error_detector_kind(self) → nessie.detectors.error_detector.DetectorKind

score(self, texts: nessie.types.StringArray, labels: nessie.types.StringArray, **kwargs) → numpy.typing.NDArray[bool]

supports_correction(self) → bool