The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.
.warning and .drift now follow one
contract across all 22 detectors: NA means the detector
cannot judge that observation yet (warm-up), FALSE that it
is active and has not flagged drift as of that observation.
Behaviour change: "kswin",
"adwin", "seed", "seqdrift2",
"fhddm", "fhddms" and the
"mddm_*" detectors used to report FALSE while
warming up and now report NA; "fhddms" and
"mddm_*", which have no warning level, now give
.warning = NA. Detections themselves are unchanged."seed" no longer has an alpha
hyperparameter: the algorithm never read it.fit(), advance(), augment()
and detect_drift() now refuse data that already has a
.warning, .drift or .phase column
instead of overwriting it, and advance() refuses a batch
whose columns differ from the baseline’s."wstd", "ftdd", "fpdd" and
"fsdd" are much faster: their per-observation tests use
closed forms that give identical p-values."kswin" with
window_size < stat_size + 2) abort with a clear message
instead of running and returning silently wrong flags.
drift_detector() also rejects a method that is
not a single string.drift_detector() gains seed: the
stochastic detectors ("kswin", "seqdrift2")
draw from a private random stream carried inside the fitted object.
Results are reproducible, independent of batching, and the session’s
global RNG is no longer advanced.drift_detector() gains keep (default
10000), the number of most recent rows retained in the
history. This bounds memory and removes the quadratic cost of row-by-row
advance(). Behaviour change: histories
longer than 10000 rows are now truncated, so augment() on a
fitted detector returns at most the last keep rows; use
keep = Inf for the previous behaviour.
keep = Inf and keep = 0 emit a warning when
the detector is specified.tidy(), glance() and print()
report running totals stored in the fitted object, so they stay exact
when the history is truncated.Depends: R (>= 4.1) and imports
utils, which "kswin" already used.credit_monitoring and sensor_monitoring, and
the vignettes (“Getting Started with deriva”, and the new
“Distribution-Based Drift Detection”) now walk through them instead of
generating data inline.drift_detector() → fit() →
advance() with augment(), tidy(),
glance(), and autoplot() generics.detect_drift() one-shot shortcut.add_prediction_error() bridge from tidymodels
workflows.sim_drift_stream() and sim_dist_stream()
for synthetic benchmarking.These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.