Change-point detection using the conditional entropy of ordinal patterns

Detecting change-point using the ordinal structure of a time series

FindAllOrdinalCP.m - detects structural change-points by using CEofOP statistic [UK15] and returns positions of all detected change-points. Ordinal patterns are computed using the fast algorithm presented in [UK13].
Input parameters of the change-point detection algorithm include:
- falseAlarmRate - assigned probability of a false alarm (erroneously detected change-point in a stationary signal). falseAlarmRate = 0.05 is acceptable in many cases.
- minCPdist - minimal distance between change-points (minimal expected length of a stationary segment). For example in EEG time series one may expect stationary segments to be at least of 1-2 seconds, which means that
minCPdist = samplingRate;
or
minCPdist = 2*samplingRate;
could be used
- order - order of ordinal patterns to be used for ordinal change-point detection (please, see [UK13, UK15] for details). Order values of 2 or 3 are recommended. If you expect that the stationary segments in the time series are longer than 20 but shorter than 100 (minCPdist > 20, but minCPdist < 100), please use you order = 2. If you expect that all stationary segments in the time series are longer than 100 (minCPdist >= 100), please use you order = 3 .
ChangePointExample - example of using FindAllOrdinalCP.
GenerateARWithChange.m, GenerateLogisticWithChange.m - generate examples of time series with structural change-points (see [UK15]).
[UK15] Unakafov, A.M. and Keller, K., 2015. Change-point detection using the conditional entropy of ordinal patterns. arXiv preprint arXiv:1510.01457.
[UK13] Unakafova, V.A., Keller, K., 2013. Efficiently measuring complexity on the basis of real-world Data. Entropy, 15(10), 4392-4415.