After successfully importing MsFPOP, we can identify changepoints within a singal by using the Ms.FPOP algorithm, through the function of the same name (MsFPOP()). For this brief demonstration, the constants \(\gamma=9\) and \(\beta=2.25\) have been calibrated based on simulations from Liehrmann and Rigaill 2023. The Gaussian signal of variance 1 is affected by 2 changepoints occuring at position 200 and 400. The mean values corresponding to the initial, intermediate, and final segments are 0, \(\sqrt{\frac{1}{6}}\), and 0 respectively.
library(MsFPOP)
set.seed(1)
beta <- 2.25
gamma <- 9
n <- 600
mu <- rep(c(0,sqrt(100/n),0), rep(200,3))
y <- rnorm(n)+mu
cp <- MsFPOP(y=y, beta=beta, alpha=gamma+beta*log(n))$changepoint
cp
#> [1] 200 402 600MsFPOP() estimates two changepoints at positions 200 and 402. It should be noted that the latter position represents the overall signal length, a value that MsFPOP() invariably returns.
References
Liehrmann, A. and Rigaill, G. Ms.Fpop: An Exact and Fast Segmentation Algorithm With a Multiscale Penalty (2023) doi:10.48550/arXiv.2303.08723.