mcp_wrapper: Bayesian formula-based changepoint regression (mcp)

In ggchangepoint: Combines Changepoint Analysis with 'ggplot2'

Bayesian formula-based changepoint regression (mcp)

Description

Wraps mcp::mcp() (Lindeløv): a Bayesian multiple-changepoint regression specified as a list of formulas, one per segment. This is the most expressive detector in the package: each segment can have its own intercept, slope, variance and autocorrelation, and the changepoints themselves get full posterior distributions rather than point estimates, summarised here as ci_lower/ci_upper on the changepoints tibble. (ggcpt_posterior() draws a per-location probability profile, which only bcp and Rbeast expose; it does not accept an mcp result.)

Usage

mcp_wrapper(
  x,
  change_in = c("mean", "slope", "var"),
  n_changepoints = 1,
  model = NULL,
  prior = list(),
  iter = 3000,
  adapt = 1000,
  chains = 3,
  seed = NULL,
  ...
)

Arguments

x	A numeric vector.
change_in	Shorthand for the segment model when model is not given: "mean" fits a change in intercept (list(y ~ 1, ~ 1)), "slope" a change in slope (list(y ~ 1 + t, ~ 0 + t)), "var" a change in residual standard deviation (list(y ~ 1, ~ 0 + sigma(1))).
n_changepoints	Number of changepoints in the shorthand model. Defaults to 1.
model	An explicit mcp model: a list of formulas. Overrides change_in and n_changepoints, and is the reason to reach for this engine at all.
prior	Optional named list of priors, passed to mcp::mcp().
iter, adapt, chains	Sampler settings, passed through.
seed	Optional seed. The seed is scoped to this call: .Random.seed is saved and restored, so a seeded call inside a simulation loop does not pin the loop's own stream.
...	Additional arguments passed to mcp::mcp().

Value

A ggcpt object. The changepoints tibble carries the posterior mean location together with ci_lower/ci_upper from the posterior quantiles, each converted from mcp's continuous cp_1, cp_2, ... to the last observation before the change (ceiling(cp) - 1, because mcp starts a segment at x >= cp), and $data$fitted holds the posterior predictive mean, so autoplot(show_ci = TRUE, show_fit = TRUE) shows both.

JAGS is a system dependency

mcp samples through JAGS, a separate program installed outside R. Having the package is not the same as being able to run it: mcp imports rjags, and on some platforms rjags installs happily and only fails when it looks for the JAGS library at run time, in which case mcp::mcp() returns a fit carrying no posterior samples, with a warning rather than an error. This wrapper checks for that and says so plainly instead of failing several frames later inside summary(). Everything else in the package works without JAGS.

References

\insertRef

lindelov2020mcpggchangepoint

See Also

Other changepoint engines: bcp_wrapper(), beast_wrapper(), bfast_wrapper(), binsegrcpp_wrapper(), bocpd_wrapper(), capa_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(), mosum_wrapper(), network_wrapper(), not_wrapper(), npmojo_wrapper(), nsp_wrapper(), ocd_wrapper(), pilliat_wrapper(), segmented_wrapper(), smuce_wrapper(), sn_wrapper(), strucchange_wrapper(), taylor_wrapper(), tguh_wrapper(), trend_wrapper(), var_wrapper(), wbs2_wrapper(), wbs_wrapper(), wbsts_wrapper()

Examples

# Not run by R CMD check: whether this works depends on a *system*
# library, and no test of installed R packages predicts that reliably --
# `rjags` can be present and still fail to find JAGS at run time.
## Not run: 
set.seed(2026)
fit <- mcp_wrapper(c(rnorm(60), rnorm(60, 4)), iter = 500, adapt = 200)
fit

## End(Not run)

ggchangepoint documentation built on Nov. 5, 2026, 9:06 a.m.