bwd: Backward procedure for the change point detection

In bwd: Backward Procedure for Change-Point Detection

Description

Implements backward procedure for detecting single or multiple change points.

Usage

bwd(y, alpha = 0.05, kmin = 3, lastkgroup = floor(0.01 * n),
  mu0 = NULL, normal = T, n.permute = 1000, h = 10)

Arguments

y	observed data
alpha	target level that detemines stopping criterion. Default is 0.05
kmin	minimum length of segements for checking possible change points
lastkgroup	We can abvoid chekcing possible change points when we have less groups than "lastkgroup" to improve computational efficiency. Default is 0.01 * n
mu0	Baseline mean value whe detecting epidemic chang points. Defalut is NULL
normal	if TRUE normal cutoff values are used, and if FALSE residual permuted cutoff values are used. Default is TRUE
n.permute	number of permutation when computing the permuted cutoff. Defalut is 1000
h	bandwidth size for variance esitimator

Value

bwd object that contains information of detected segments and significance levels

Author(s)

Seung Jun Shin, Yicaho Wu, Ning Hao

References

Shin, Wu, and Hao (2018+) A backward procedure for change-point detection with applications to copy number variation detection, arXiv:1812.10107.

See Also

plot.bwd

Examples

# simulated data
set.seed(1)
n <- 1000
L <- 10

mu0 <- -0.5

mu <- rep(mu0, n)
mu[(n/2 + 1):(n/2 + L)] <- mu0 + 1.6
mu[(n/4 + 1):(n/4 + L)] <- mu0 - 1.6
y <- mu + rnorm(n)
alpha <- c(0.01, 0.05)

# BWD
obj1 <- bwd(y, alpha = alpha)

# Modified for epidemic changes with a known basline mean, mu0.
obj2 <- bwd(y, alpha = alpha, mu0 = 0)

par(mfrow = c(2,1))
plot(obj1, y)
plot(obj2, y)

bwd documentation built on May 2, 2019, 6:45 a.m.