dataeval.detectors.drift.DriftOutput¶

class dataeval.detectors.drift.DriftOutput¶

Output class for univariate drift detectors.

Extends DriftBaseOutput with feature-level (per-pixel) drift information. Used by Kolmogorov-Smirnov, Cramér-von Mises, and uncertainty-based drift detectors that analyze each feature independently.

drifted¶

Overall drift prediction after multivariate correction.

Type:¶

bool

threshold¶

Corrected threshold after Bonferroni or FDR correction for multiple testing.

Type:¶

float

p_val¶

Mean p-value across all features, between 0 and 1. For descriptive purposes only; individual feature p-values are used for drift detection decisions. Can appear high even when drifted=True if only a subset of features show drift.

Type:¶

float

distance¶

Mean test statistic across all features, always >= 0.

Type:¶

float

feature_drift¶

Boolean array indicating which features (pixels) show drift. Shape matches the number of features in the input data.

Type:¶

NDArray[bool]

feature_threshold¶

Uncorrected p-value threshold used for individual feature testing. Typically the original p_val before multivariate correction.

Type:¶

float

p_vals¶

P-values for each feature, all values between 0 and 1. Shape matches the number of features in the input data.

Type:¶

NDArray[np.float32]

distances¶

Test statistics for each feature, all values >= 0. Shape matches the number of features in the input data.

Type:¶

NDArray[np.float32]

Notes

Feature-level analysis enables identification of specific pixels or regions that contribute most to detected drift, useful for interpretability.

data()¶

The output data as a dictionary.

Return type:¶

dict[str, Any]

meta()¶

Metadata about the execution of the function or method for the Output class.

Return type:¶

ExecutionMetadata