Second, "Geometric-based pruning rules for change point detection in multiple independent time series" by Liudmila Pishchagina, Guillem Rigaill, and Vincent Runge is available at doi.org/10.57750/9vvx-eq57 and complemented with R/C++ code at https://github.com/lpishchagina/GeomFPOP
The paper focuses on finding an unknown number of change points in multiple independent time-series, assuming the change points occur simultaneously in all series. The authors focus on dynamic programming algorithms and propose GeomFPOP (Geometric Functional Pruning Optimal Partitioning). GeomFPOP uses geometric pruning rules, where the state-of-the-art PELT method uses inequality-based pruning rules. Among other things, the authors show that GeomFPOP can be significantly faster when the number of change points is small with respect to the size of the time series.
#reproducibility #openScience #openAccess #openSource #rStats #changePoint
