drift#

Drift detectors identify if the statistical properties of the data has changed.

Detector Classes#

DriftCVM(x_ref[, p_val, x_ref_preprocessed, ...])

Drift detector employing the Cramér-von Mises (CVM) Drift Detection test.

DriftKS(x_ref[, p_val, x_ref_preprocessed, ...])

Drift detector employing the Kolmogorov-Smirnov (KS) distribution test.

DriftMMD(x_ref, p_val, x_ref_preprocessed, ...)

Maximum Mean Discrepancy (MMD) Drift Detection algorithm using a permutation test.

DriftUncertainty(x_ref, model[, p_val, ...])

Test for a change in the number of instances falling into regions on which the model is uncertain.

Supporting Classes#

Kernels#

Kernels are used to map non-linear data to a higher dimensional space.

kernels.GaussianRBF([sigma, init_sigma_fn, ...])

Gaussian RBF kernel: k(x,y) = exp(-(1/(2*sigma^2)||x-y||^2).

Update Strategies#

Update strategies inform how the drift detector classes update the reference data when monitoring for drift.

updates.LastSeenUpdate(n)

Updates reference dataset for drift detector using last seen method.

updates.ReservoirSamplingUpdate(n)

Updates reference dataset for drift detector using reservoir sampling method.

Output Classes#

DriftOutput(is_drift, threshold, p_val, ...)

Output class for DriftCVM, DriftKS, and DriftUncertainty drift detectors

DriftMMDOutput(is_drift, threshold, p_val, ...)

Output class for DriftMMD drift detector