| Title: | Joint Change Point Detection |
|---|---|
| Description: | Procedures for joint detection of changes in both expectation and variance in univariate sequences. Performs a statistical test of the null hypothesis of the absence of change points. In case of rejection performs an algorithm for change point detection. Reference - Bivariate change point detection - joint detection of changes in expectation and variance, Scandinavian Journal of Statistics, DOI 10.1111/sjos.12547. |
| Authors: | Michael Messer [aut, cre] |
| Maintainer: | Michael Messer <[email protected]> |
| License: | GPL-3 |
| Version: | 1.2 |
| Built: | 2024-12-15 07:30:32 UTC |
| Source: | CRAN |
Joint change point detection - expectation and variance - via bivariate moving sum statistics
jcp(x, H = NA, q = NA, alpha = 0.05, sim = 1000, region = "square")jcp(x, H = NA, q = NA, alpha = 0.05, sim = 1000, region = "square")
x |
numeric vector. Input sequence of random variables. |
H |
NA or numeric vector. Window set. If NA (default), then H is automatically set. If not NA, then H must an increasing vector of positive integers with maximum =< length(x)/2. |
q |
NA or numeric value. Rejection threshold. If NA (default), then the rejection boundary is derived in simulations (from Gaussian process limit) according to sim and alpha. If not NA, then q is considered predefined and must be set a postive real number. |
alpha |
numeric value. Significance level. Must be in (0,1), default = 0.05. In case of predefined q, alpha is set to NA. |
sim |
numeric value. Number of simulations of limit process for approximation of q. Must be positive integer, default = 1000. In case of predefined q, sim is set to NA. |
region |
character string. Defines rejection region, default = "square". Must be chosen either "square", "circle" or "ellipse". |
invisible list
changepoints |
detected change points (increasingly ordered) |
mean_sd |
matrix of estimated means and standard deviations |
M |
test statistic |
q |
rejection threshold |
H |
window set |
sim |
number of simulations of the limit process (approximation of q) |
alpha |
significance level |
region |
rejection region |
method |
derivation of threshold q, either asymptotic or predefined |
x |
input sequence |
EVrho |
list containing the auxiliary processes E, V and correlation rho, for each element of H one list entry |
CP_meta |
matrix containing meta information of estimation. Estimated change points (increasingly ordered), responsible window h, components E, V and rho of joint statistic at estimated change points (regarding responsible window) |
SFA |
detected change points of single filter algorithms |
Michael Messer
Michael Messer (2021) Bivariate change point detection - joint detection of changes in expectation and variance, Scandinavian Journal of Statistics, DOI 10.1111/sjos.12547.
# Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) summary(result) plot(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) summary(result2) plot(result2)# Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) summary(result) plot(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) summary(result2) plot(result2)
Plot method for class 'jcp'
## S3 method for class 'jcp' plot(x, cex = 1, cex.main = 1, ...)## S3 method for class 'jcp' plot(x, cex = 1, cex.main = 1, ...)
x |
object of class jcp |
cex |
numeric, global sizes in plot |
cex.main |
numeric, size of titles |
... |
additional arguments |
No return value, called for side effects
Michael Messer
Michael Messer (2021) Bivariate change point detection - joint detection of changes in expectation and variance, Scandinavian Journal of Statistics, DOI 10.1111/sjos.12547.
# Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) summary(result) plot(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) summary(result2) plot(result2)# Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) summary(result) plot(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) summary(result2) plot(result2)
Summary method for class 'jcp'
## S3 method for class 'jcp' summary(object, ...)## S3 method for class 'jcp' summary(object, ...)
object |
object of class jcp |
... |
additional arguments |
No return value, called for side effects
Michael Messer
Michael Messer (2021) Bivariate change point detection - joint detection of changes in expectation and variance, Scandinavian Journal of Statistics, DOI 10.1111/sjos.12547.
#' # Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) plot(result) summary(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) plot(result2) summary(result2)#' # Normal distributed sequence with 3 change points at # c1=250 (change in expectation), # c2=500 (change in variance) and # c3=750 (change in expectation and variance) set.seed(0) m <- c(8,10,10,3); s <- c(4,4,10,5) x <- rnorm(1000, mean=rep(m,each=250), sd=rep(s,each=250)) result <- jcp(x) plot(result) summary(result) # Set additional parameters (window set) result2 <- jcp(x,H=c(80,160,240)) plot(result2) summary(result2)