SMUCE / HSMUCE wrapper: multiscale changepoint inference

Wraps stepR::stepFit() for the Simultaneous MUltiscale Changepoint Estimator (SMUCE) of Frick, Munk and Sieling (2014) and its heterogeneous extension HSMUCE (Pein, Sieling and Munk, 2017). SMUCE estimates a step function subject to a simultaneous multiscale test at level alpha; the level bounds the probability of over-estimating the number of changepoints, and the fit delivers confidence intervals for every changepoint location, which populate the ci_lower/ci_upper columns of the result and render via autoplot(show_ci = TRUE) or geom_cpt_ci().

Usage

smuce_wrapper(
  x,
  alpha = 0.5,
  family = c("gauss", "hsmuce", "poisson", "binomial"),
  ...
)

Arguments

x

A numeric vector.

alpha

Significance level of the multiscale test in \((0, 1)\); smaller values yield more conservative (fewer-changepoint) fits. Defaults to 0.5, the upstream recommendation for estimation.

family

Noise model: "gauss" (SMUCE, homogeneous Gaussian noise), "hsmuce" (HSMUCE, segment-wise variance), "poisson" (counts: a change in the rate) or "binomial" (0/1 outcomes: a change in the probability). Defaults to "gauss". The two count families run stepR::smuceR(), the multiscale estimator for exponential families, with the same level and the same jump intervals; cpt_detect(x, method = "smuce", family = "poisson") reaches them. stepR documents smuceR() as deprecated but working, and as the only route to these families until stepFit() gains them; should it be removed, the error says so and names the methods that fit the same family. With alpha given, smuceR() takes its threshold from the asymptotic null distribution rather than a simulated one. The remaining stepR families ("jsmurf", "mDependentPS", ...) all require a filter or covariance specification; call stepR::stepFit() directly for those. "hsmuce" additionally refuses a series whose point-to-point variation lies more than about seven orders of magnitude below its own scale: a globally flat series, or a step whose segments are numerically constant, as cpt_simulate(sd = 0) produces once any rounding is added. stepR's heterogeneous variance estimator aborts the R session on such input rather than raising an error, so it cannot be caught. "gauss" handles the whole range.

...

Additional arguments passed to stepR::stepFit() (stepR::smuceR() for the count families).

Value

A ggcpt object. The changepoints tibble carries ci_lower/ci_upper (confidence interval for each changepoint location) and the data tibble carries the SMUCE step fit in its fitted column.

References

Frick K, Munk A, Sieling H (2014). “Multiscale change point inference.” Journal of the Royal Statistical Society: Series B, 76(3), 495–580.

Pein F, Sieling H, Munk A (2017). “Heterogeneous change point inference.” Journal of the Royal Statistical Society: Series B, 79(4), 1207–1227.

See also

Other changepoint engines: bcp_wrapper(), beast_wrapper(), bfast_wrapper(), binsegrcpp_wrapper(), bocpd_wrapper(), cpm_wrapper(), cpop_wrapper(), cpt_wrapper(), decafs_wrapper(), ecp_wrapper(), envcpt_wrapper(), esac_wrapper(), fabisearch_wrapper(), fastcpd_wrapper(), fcov_wrapper(), fmean_wrapper(), fpop_wrapper(), geomcp_wrapper(), hdcov_wrapper(), hdreg_wrapper(), idetect_wrapper(), inspect_wrapper(), kcp_wrapper(), kwc_wrapper(), mcp_wrapper(), mosum_wrapper(), network_wrapper(), not_wrapper(), npmojo_wrapper(), nsp_wrapper(), ocd_wrapper(), pilliat_wrapper(), segmented_wrapper(), sn_wrapper(), strucchange_wrapper(), taylor_wrapper(), tguh_wrapper(), trend_wrapper(), var_wrapper(), wbs2_wrapper(), wbs_wrapper(), wbsts_wrapper()

Examples

# \donttest{
set.seed(2026)
x <- c(rnorm(100), rnorm(100, 3))
res <- smuce_wrapper(x)
res$changepoints
#> # A tibble: 1 × 4
#>      cp cp_value ci_lower ci_upper
#>   <int>    <dbl>    <int>    <int>
#> 1   100    0.369       99      101
ggplot2::autoplot(res, show_ci = TRUE)

# }