Wraps kcpRS::kcpRS() (Cabrieto et al., 2018; the KCP framework of
Arlot, Celisse and Harchaoui, 2019). The data are mapped to a running
statistic (mean, variance, autocorrelation, or correlation) computed on a
sliding window, and a Gaussian-kernel change point analysis with a
permutation significance test is run on the statistic. Detecting changes
in running correlations or variances captures higher-order changes that
mean-based methods miss. Multivariate input (matrix or data frame) is
supported.
Usage
kcp_wrapper(
x,
running_stat = c("mean", "var", "autocorr", "corr"),
wsize = 25,
nperm = 1000,
kmax = 10,
alpha = 0.05,
seed = NULL,
...
)Arguments
- x
A numeric vector, matrix, or data frame (columns are variables).
- running_stat
Which running statistic to monitor:
"mean","var","autocorr", or"corr"(correlation requires at least two columns). Defaults to"mean".- wsize
Sliding window size for the running statistic. Defaults to
25.- nperm
Number of permutations for the significance test. Defaults to
1000.- kmax
Maximum number of changepoints considered. Defaults to
10.- alpha
Significance level of the permutation test. Defaults to
0.05.- seed
Optional seed for reproducibility of the permutation test.
- ...
Additional arguments passed to
kcpRS::kcpRS().
Value
A ggcpt object. Reported locations refer to the centre of
the sliding window in which the change occurs.
References
Arlot S, Celisse A, Harchaoui Z (2019). “A kernel multiple change-point algorithm via model selection.” Journal of Machine Learning Research, 20(162), 1–56.
Cabrieto J, Adolf J, Tuerlinckx F, Kuppens P, Ceulemans E (2018). “Detecting long-lived autodependency changes in a multivariate system via change point detection and regime switching models.” Scientific Reports, 8, 15637.