Package index • deriva

Specifying and fitting detectors

Choose a detector from the catalogue of 22 methods, fit it on a baseline period, and advance it over new observations.

drift_detector(): Specify a drift detector
fit(<drift_detector>): Fit a drift detector on a baseline period
advance(): Advance a fitted drift detector over a new batch
detect_drift(): Detect drift in a signal column (one-shot shortcut)

Inspecting results

Tidy accessors for a fitted detector and its stream of flags.

augment(<drift_detector_fit>): Annotated observations from a fitted drift detector
tidy(<drift_detector_fit>): Drift points of a fitted detector
glance(<drift_detector_fit>): One-row summary of a fitted detector
autoplot(<drift_detector_fit>): Plot the monitored signal with drift markings

Working with tidymodels

Turn model predictions into the per-observation error signal detectors consume.

add_prediction_error(): Build a drift signal from model predictions

Simulating streams

Synthetic streams with known change points, for benchmarking and teaching.

sim_drift_stream(): Simulate a binary error stream with a known drift point
sim_dist_stream(): Simulate a continuous stream with a known distribution-shift point

Package

deriva deriva-package: deriva: Tidy Drift Detection for Monitored Machine Learning Models
reexports augment fit tidy glance: Objects exported from other packages