kcpRS {kcpRS}

KCP on the running statistics

Description

Given a user-specified function RS_fun to compute the running statistics (see runMean, runVar, runAR and runCorr), a KCP permutation test (see permTest) is first implemented to test whether there is at least one significant change point, then through model selection most optimal number of change points is chosen.

Usage

kcpRS(
  data,
  RS_fun,
  RS_name,
  wsize = 25,
  nperm = 1000,
  Kmax = 10,
  alpha = 0.05,
  varTest = FALSE,
  ncpu = 1
)

## S3 method for class 'kcpRS'
plot(x, ...)

## S3 method for class 'kcpRS'
print(x, kcp_details = TRUE, ...)

## S3 method for class 'kcpRS'
summary(object, ...)

Arguments

data

data N x v dataframe where N is the number of time points and v the number of variables

RS_fun

Running statistics function: Should require wsize and wstep as input and return a dataframe of running statistics as output. The rows of this dataframe should correspond to the windows and the columns should correspond to the variable(s) on which the running statistics were computed.

RS_name

Name of the monitored running statistic.

wsize

Window size

nperm

Number of permutations used in the permutation test

Kmax

Maximum number of change points desired

alpha

Significance level of the permutation test

varTest

If set to FALSE, only the variance DROP test is implemented, and if set to TRUE, both the variance test and the variance DROP tests are implemented.

ncpu

number of cpu cores to use

x

An object of the type produced by kcpRS

...

Further plotting arguments.

kcp_details

If TRUE, then the matrix of optimal change points solutions given k is displayed. If FALSE, then this output is suppressed.

object

An object of the type produced by kcpRS_workflow

Value

RS_name

Name indicated for the monitored running statistic

RS

Dataframe of running statistics with rows corresponding to the time window and columns corresponding to the (combination of) variable(s) on which the running statistics were computed

wsize

Selected window size

varTest

Selected choice of implementation for varTest

nperm

Selected number of permutations

alpha

Selected significance level of the permutation test

subTest_alpha

Significance level of each subtest. If varTest=0, subTest_alpha is equal to alpha since only the variance drop test is implemented. If varTest=1, subTest_alpha=alpha/2 since two subtests are carried out and Bonferonni correction is applied.

BestK

Optimal number of change points based on grid search

changePoints

Change point location(s)

p_var_test

P-value of the variance test

p_varDrop_test

P-value of the variance drop test

CPs_given_K

A matrix comprised of the minimized variance criterion Rmin and the optimal change point location(s) for each k from 1 to Kmax

changePoints_scree_test

Optimal number of change points based on scree test

scree_test

A matrix comprised of the scree values for each k from 1 to Kmax-1

medianK

Median Euclidean distance between all pairs of running statistics

References

Cabrieto, J., Tuerlinckx, F., Kuppens, P., Wilhelm, F., Liedlgruber, M., & Ceulemans, E. (2018). Capturing correlation changes by applying kernel change point detection on the running correlations. Information Sciences, 447, 117-139. doi:10.1016/j.ins.2018.03.010

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, 1-15. doi:10.1038/s41598-018-33819-8

Cabrieto, J., Meers, K., Schat, E., Adolf, J. K., Kuppens, P., Tuerlinckx, F., & Ceulemans, E. (2022). kcpRS: An R package for performing kernel change point detection on the running statistics of multivariate time series. Behavior Research Methods, 54, 1092-1113. doi:10.3758/s13428-021-01603-8

Examples

phase1=cbind(rnorm(50,0,1),rnorm(50,0,1)) #phase1: Means=0
phase2=cbind(rnorm(50,1,1),rnorm(50,1,1)) #phase2: Means=1
X=rbind(phase1,phase2)
res=kcpRS(data=X,RS_fun=runMean,RS_name="Mean",wsize=25,
nperm=1000,Kmax=10,alpha=.05,varTest=FALSE,ncpu=1)

summary(res)
plot(res)