Sequential change point model wrapper (CPM)

Wraps cpm::processStream() (Ross, 2015) for distribution-free sequential changepoint detection via repeated two-sample tests (Mann-Whitney for location, Mood for scale, Lepage, Kolmogorov-Smirnov and Cramer-von-Mises for general changes, and parametric Student/Bartlett/GLR variants). Although the engine is designed for streams, it is run here over the full series in one pass, mimicking online monitoring with average run length arl0.

Usage

cpm_wrapper(x, cpm_type = "Mann-Whitney", arl0 = NULL, startup = 20, ...)

Arguments

x

A numeric vector.

cpm_type

Test statistic, passed to cpm::processStream() as cpmType. Distribution-free: "Mann-Whitney" (location, the default), "Mood" (scale), "Lepage", "Kolmogorov-Smirnov", "Cramer-von-Mises". Parametric: "Student", "Bartlett", "GLR" (Gaussian), "Exponential" (positive data), "FET" (Fisher's exact test, for 0/1 Bernoulli data, with a smoothing lambda passed through ...: 0.1, cpm's default, or 0.3).

arl0

Target in-control average run length (how many observations, on average, before a false alarm). cpm ships thresholds only for a fixed grid (100, 200, 300, 370, 400, 500, 600, 700, 800, 900, 1000, 2000, 3000, 4000, 5000, 6000, 7000, 8000, 9000, 10000, 20000, 30000, 40000 and 50000), and any other value is refused, because the engine answers it by printing an error and reporting no changepoints. The grid is the same for every cpm_type, and 50000 is the ceiling: a long series cannot be given an arl0 proportional to its length indefinitely.

When NULL (the default) it scales with the series: the smallest grid value of at least \(5n\), capped at 50000. A sequential test run over a whole series raises about \(n / \mathrm{arl0}\) false alarms, so the streaming default of 500 meant 36 changepoints on 10,000 observations with 4 real ones, and 350 on 100,000; \(5n\) turned the 36 into exactly the 4. The expected count is recorded as $diagnostics$expected_false_positives. Pass arl0 = 500 for the 0.5.0 behaviour, and use cpt_monitor() for genuine streaming, whose default is unchanged.

startup

Number of observations after each restart before monitoring begins. Defaults to 20.

...

Additional arguments passed to cpm::processStream().

Value

A ggcpt object. The changepoints tibble carries a detection_time column: the index at which the sequential test flagged each change (always later than the estimated location).

References

Ross GJ (2015). “Parametric and nonparametric sequential change detection in R: The cpm package.” Journal of Statistical Software, 66(3), 1–20.

See also

Other changepoint engines: bcp_wrapper(), beast_wrapper(), bfast_wrapper(), binsegrcpp_wrapper(), bocpd_wrapper(), cpop_wrapper(), cpt_wrapper(), decafs_wrapper(), ecp_wrapper(), envcpt_wrapper(), esac_wrapper(), fabisearch_wrapper(), fastcpd_wrapper(), fcov_wrapper(), fmean_wrapper(), fpop_wrapper(), geomcp_wrapper(), hdcov_wrapper(), hdreg_wrapper(), idetect_wrapper(), inspect_wrapper(), kcp_wrapper(), kwc_wrapper(), mcp_wrapper(), mosum_wrapper(), network_wrapper(), not_wrapper(), npmojo_wrapper(), nsp_wrapper(), ocd_wrapper(), pilliat_wrapper(), segmented_wrapper(), smuce_wrapper(), sn_wrapper(), strucchange_wrapper(), taylor_wrapper(), tguh_wrapper(), trend_wrapper(), var_wrapper(), wbs2_wrapper(), wbs_wrapper(), wbsts_wrapper()

Examples

res <- cpm_wrapper(c(rnorm(100), rnorm(100, 3)))
res$changepoints
#> # A tibble: 1 × 3
#>      cp cp_value detection_time
#>   <int>    <dbl>          <int>
#> 1   100   -0.125            105