VARDetect (version 0.1.8)

detection_check: Function for detection performance check

Description

Function for detection performance check

Usage

detection_check(pts.final, brk, nob, critval = 5)

Value

a matrix of detection summary results, including the absolute error, selection rate and relative location. The absolute error of the locations of the estimated break points is defined as \({error}_j =|\tilde{t}_j^f - t_j|\), \(j = 1,\dots, m_0\).

Arguments

pts.final

a list of estimated change points

brk

the true change points

nob

length of time series

critval

critical value for selection rate. Default value is 5. Specifically, to compute the selection rate, a selected break point is counted as a ``success'' for the \(j\)-th true break point, \(t_j\), if it falls in the interval \([t_j - {(t_{j} - t_{j-1})}/{critval}, t_j + {(t_{j+1} - t_{j})}/{critval}]\), \(j = 1,\dots, m_0\).

Examples

# an example of 10 replicates result
set.seed(1)
nob <- 1000
brk <- c(333, 666, nob+1)
cp.list <- vector('list', 10)
for(i in 1:10){
    cp.list[[i]] <-  brk[1:2] + sample(c(-50:50),1)
}
# some replicate fails to detect all the change point
cp.list[[2]] <- cp.list[[2]][1]
cp.list[4] <- list(NULL)      # setting 4'th element to NULL.
# some replicate overestimate the number of change point
cp.list[[3]] <- c(cp.list[[3]], 800)
cp.list
res <- detection_check(cp.list, brk, nob, critval = 5)
res
# use a stricter critical value
res <- detection_check(cp.list, brk, nob, critval = 10)
res