Changepoint detectionAllen B. DowneyThis work has been superseded. The probability it computes, that each index is the most recent changepoint, is the run-length posterior of Bayesian online changepoint detection (Adams and MacKay, arXiv:0710.3742, 2007), which computes it exactly and efficiently; see also Fearnhead (Statistics and Computing, 2006). The implementation used for the results in the paper below contained errors. A corrected implementation, tests, and a description of the errors are in the ChangePoint repository on GitHub. The material below is kept as an archive. Changepoint detection in network measurementsAbstractWe propose an algorithm for simultaneously detecting and locating changepoints in a time series, and a framework for predicting the distribution of the next point in the series. The kernel of the algorithm is a system of equations that computes, for each index $i$, the probability that the last (most recent) change point occurred at $i$. We evaluate this algorithm by applying it to the change point detection problem and comparing it to the generalized likelihood ratio (GLR) algorithm. We find that our algorithm is as good as GLR, or better, over a wide range of scenarios, and that the advantage increases as the signal-to-noise ratio decreases.PapersThe paper is on arXiv as arXiv:0812.1237. The 2008 draft is also available here in Postscript, gzipped Postscript and PDF.SlidesHere are the slides I used when I presented this work. SoftwareSee the ChangePoint repository, which includes the original code and a corrected implementation. |