Geometrically-inspired multivariate changepoint wrapper (geomcp)

Wraps changepoint.geo::geomcp() (Grundy, Killick and Mihaylov, 2020): each multivariate observation is mapped to its distance from, and angle to, a reference point, and univariate PELT is run on the two mapped series. Distance changes capture shifts in magnitude, angle changes capture shifts in orientation/correlation structure.

Usage

geomcp_wrapper(
  x,
  penalty = "MBIC",
  mapping = c("both", "distance", "angle"),
  ...
)

Arguments

x: A numeric matrix or data frame with one row per time point.
penalty: Penalty for the univariate PELT runs (a changepoint-style character penalty). Defaults to "MBIC".
mapping: Which mapped series' changepoints to report: "both" (union, default), "distance", or "angle".
...: Additional arguments passed to changepoint.geo::geomcp().

Value

A ggcpt object whose changepoints tibble carries a mapping column ("distance" or "angle"; a location found in both is labelled "both").

References

Grundy T, Killick R, Mihaylov G (2020). “High-dimensional changepoint detection via a geometrically inspired mapping.” Statistics and Computing, 30, 1155–1166.

Examples

set.seed(2026)
X <- rbind(matrix(rnorm(100 * 4), 100), matrix(rnorm(100 * 4, 2), 100))
res <- geomcp_wrapper(X)
res$changepoints
#> # A tibble: 2 × 3
#>      cp cp_value mapping 
#>   <int>    <dbl> <chr>   
#> 1    99   -1.08  angle   
#> 2   100    0.369 distance