jointseg (version 1.0.2)

segmentByRBS: Recursive Binary Segmentation (low-level)

Description

Low-level function for multivariate Recursive Binary Segmentation (RBS)

Usage

segmentByRBS(Y, K, minRegionSize = 2, verbose = FALSE)

Arguments

Y

A n*p signal to be segmented

K

The number of change points to find

minRegionSize

Regions with less than minRegionSize are not split

verbose

A logical value: should extra information be output ? Defaults to FALSE.

Value

A list with elements:

bkp

A vector of K estimated breakpoint positions, sorted by order of appearance

rse

the residual squared error (RSE) for the successive segmentations

gain

The gain provided by each breakpoints in terms of difference between RSE

Details

This function recrusively looks for the best candidate change point according to binary segmentation. This is the low-level function. It is generally advised to use the wrapper doRBS which also works on data frames and has a convenient argument stat.

See jointSeg for combining recursive binary segmentation with pruning by dynamic programming (pruneByDP).

See PSSeg for segmenting genomic signals from SNP arrays.

Each dimension of the original signal is scaled before segmentation, using estimateSd.

References

Gey, S., & Lebarbier, E. (2008). Using CART to Detect Multiple Change Points in the Mean for Large Sample. http://hal.archives-ouvertes.fr/hal-00327146/

See Also

PSSeg, jointSeg, doRBS, pruneByDP

Examples

# NOT RUN {
p <- 2
trueK <- 10
len <- 1e4
sim <- randomProfile(len, trueK, 1, p)
Y <- sim$profile
K <- 2*trueK
res <- segmentByRBS(Y, K)
getTpFp(res$bkp, sim$bkp, tol=10, relax = -1)   ## true and false positives

cols <- rep(2, K)
cols[1:trueK] <- 3
par(mfrow=c(p,1))
for (ii in 1:p) {
    plot(Y[, ii], pch=19, cex=0.2)
    abline(v=res$bkp[1:trueK], col= cols)
    abline(v=sim$bkp, col=8, lty=2)
}

## NA:s in one dimension at a true breakpoint
jj <- sim$bkp[1]
Y[jj-seq(-10, 10), p] <- NA
res2 <- segmentByRBS(Y, K)
getTpFp(res2$bkp, sim$bkp, tol=10, relax = -1)   ## true and false positives

## NA:s in both dimensions at a true breakpoint
Y[jj-seq(-10, 10), ] <- NA
res3 <- segmentByRBS(Y, K)
getTpFp(res3$bkp, sim$bkp, tol=10, relax = -1)   ## true and false positives

# }