NEWS

CptNonPar 0.3.2 (2026-01-20)

Fixed a bug introduced in 0.3.0 where the largest importance scores were not being used in the multi-lag sequential mergin procedure.

By default, the data is now centered and scaled before change point detection is applied, which improves empirical performance. This can be turned off by setting the new argument scale.data = FALSE in the functions np.mojo(), np.mojo.multilag(), and multiscale.np.mojo().

The associated paper is now accepted for publication in Biometrika: see doi:10.1093/biomet/asaf024 for full details.
Updated multiscale.np.mojo() function so that returned cpts are given in time order.
The p-values returned by np.mojo(), np.mojo.multilag(), and mulsticale.np.mojo() have been replaced by importance scores: when the bootstrap is used, these are essentially one minus the p-value. Larger scores give more evidence of a change point.
Fixed error that occurs when using the manual threshold and the np.mojo.multilag() function, thanks to Chuanyang Zhang for spotting this.
You can now use different manual thresholds for different lags: for use in np.mojo.multilag(), you can supply a vector, whilst for mulsticale.np.mojo(), you can supply a list of vectors.
Added package level documentation: see ?CptNonPar.

Added a new function multiscale.np.mojo(), this function performs multiscale, multi-lag nonparametric change point detection using a set of bandwidths (as opposed to a single bandwidth as in np.mojo.mulilag()).
Updated parallelisation code in np.mojo() to avoid unnecessary call to closeAllConnections(), thanks to Henrik Bengtsson for highlighting this.
Changed uses of itemize in .Rd files to comply with CRAN check.