mscp {mscp}

mscp

Description

Multiscale change point detection via gradual bandwidth adjustment in moving sum processes. A method for the detection of changes in the expectation in univariate sequences.

Usage

mscp(x, delta = 20, g = 20, kappa = NA, alpha = 0.01, sim = 500)

Arguments

x

numeric vector. Input sequence of random variables.

delta

integer >=2. Default = 20. Minimal window considered.

g

integer >=1. Default = 20. Spacing between starting points.

kappa

NA or positive real number. Default = NA. Breaking threshold. If NA, then kappa is derived in simulations, using alpha and sim

alpha

numeric in (0,1). Default = 0.01. Significance level, i.e., sets kappa as (1-alpha)-quantile of maximum of Gaussian process limit.

sim

integer >=1. Default = 500. Number of simulations for kappa.

Value

invisible list

cp

detected change points (ordered according to detection)

mean_sd

matrix of estimated means and standard deviations

path

list containing matrices, each matrix describing the path of a detected change point. First column: t-value, second column: h-value, third column: D-value (statistic), first row: starting values, last row: end values

S

matrix of possible starting values. First column: t-value, second column: h-value, third column: D-value (statistic), fourth column: step when cut out

x

input sequence

delta

minimal window size

g

spacing between starting points

kappa

threshold

Author(s)

Tijana Levajkovic and Michael Messer

References

Multiscale change point detection via gradual bandwidth adjustment in moving sum processes (2021+), Tijana Levajkovic and Michael Messer

See Also

plot.mscp, summary.mscp

Examples

set.seed(1)
Tt <- 1000
cp <- c(250,500,600,650,750)
mu <- c(2,3,6,9,12,15)
sd <- c(1,1,2,1,2,1)
m  <- rep(mu,diff(c(0,cp,Tt))) 
s  <- rep(sd,diff(c(0,cp,Tt)))    
x  <- rnorm(Tt,m,s)
result <- mscp(x,kappa=4.77) # kappa set manually
# result <- mscp(x) # kappa derived in simulations
summary(result)
plot(result)