The package uses fitting axes-aligned rectangles to a time series in order to find structural breaks. The algorithm enclose the time series in a number of axes-aligned rectangles and tries to minimize their area and number. As these are conflicting aims, the user has to specify a parameter alpha in [0.0,1.0]. Values close to 0 result in more breakpoints, values close to 1 in fewer. The left edges of the rectangles are the breakpoints. The package supplies two methods, computeBreakPoints(series,alpha) which returns the indices of the break points and computeRectangles(series,alpha) which returns the rectangles. The algorithm is randomised; it uses a genetic algorithm. Therefore, the break point sequence found can be different in different executions of the method on the same data, especially when used on longer series of some thousand observations. The algorithm uses a range-tree as background data structure which makes i very fast and suited to analyse series with millions of observations. A detailed description can be found in Paul Fischer, Astrid Hilbert, Fast detection of structural breaks, Proceedings of Compstat 2014.

Author | Paul Fischer [aut, cre, cph], Astrid Hilbert [ctb, cph] |

Date of publication | 2014-07-20 21:17:32 |

Maintainer | Paul Fischer <pafi@dtu.dk> |

License | GPL-2 |

Version | 0.26 |

http://www2.imm.dtu.dk/~pafi/StructBreak/index.html |

SBRect

SBRect/inst

SBRect/inst/java

SBRect/inst/java/sbrect_jr.jar

SBRect/NAMESPACE

SBRect/R

SBRect/R/onLoad.R
SBRect/R/computeRectangles.R
SBRect/R/computeBreakPoints.R
SBRect/MD5

SBRect/DESCRIPTION

SBRect/man

SBRect/man/computeBreakPoints.Rd
SBRect/man/computeRectangels.Rd
Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.