LoginSign Up
← Back to all sparks
C

CptNonPar

INFRA · APIS
Velocity0.0

Nonparametric Change Point Detection for Multivariate Time Series

github.com/euanmcgonigle/cptnonpar ↗

Nonparametric change point detection swaps p-values for importance scores.

change-point-detectionnonparametricdefaultspreprocessingstatistics
◆Current state
CptNonPar implements nonparametric MOJO change point detection for possibly multivariate, serially dependent data, through single-lag, multi-lag and multiscale entry points. Recent releases concern how results are reported and how data is preprocessed rather than new detection machinery. The underlying method was accepted at Biometrika during the 0.3.0 cycle.
◆Where it's heading
The package is tightening the statistical interface it exposes: p-values gave way to importance scores across all three detection functions, manual thresholds became specifiable per lag, and the latest release makes centring and scaling the default preprocessing step. Each change folds a decision the user previously had to make into the package itself.
◆Prediction
Expect further work on defaults and reporting around the existing MOJO estimators rather than a new detection method.

◆Recent moves

  1. 8mo ago

    Data centred and scaled by default before detection

    Centring and scaling now run before detection by default, with scale.data = FALSE to opt out. Existing scripts will return different results, though the estimator itself is unchanged.

    View source ↗
  2. 1y ago

    Importance scores replace p-values; per-lag manual thresholds

    The reporting interface changes: p-values become importance scores across np.mojo(), np.mojo.multilag() and multiscale.np.mojo(), change points come back in time order, and manual thresholds can now vary by lag. The detection method underneath is the one now published in Biometrika.

    View source ↗
  3. 2y ago

    Paper link updated for CRAN checks

    A description link updated to satisfy a CRAN check, with no functional change to detection.

    View source ↗
  4. 3y ago

    Description field and example cleanups

    Description and example touch-ups on the early CRAN releases, with no effect on detection behaviour.

    View source ↗