Skip to contents

Glance at a ggcpt object

Source: R/broom-methods.R
glance.ggcpt.Rd

Returns a one-row summary of a changepoint detection result.

Usage

# S3 method for class 'ggcpt'
glance(x, ...)

Arguments

x

A ggcpt object.

...

Additional arguments (ignored).

Value

A one-row tibble with columns: n, n_changepoints, method, change_in, penalty_type, penalty_value, cp_convention, total_cost (NA when the engine does not expose a cost), runtime (elapsed seconds when measured by cpt_detect(), otherwise NA).

Details

total_cost is reported on whatever scale the engine itself uses, so it is meaningful when comparing penalties within one method and not when comparing one method against another. For the changepoint engines it is the unpenalised \(-2\log L\) of the chosen segmentation. Four cases there are NA rather than filled with a number that would not mean the same thing:

  • "binseg" and "segneigh", whose cpt.range fits report the raw within-segment cost instead: for one and the same segmentation that is 219.7 where a PELT fit reports 659.9;

  • "np", because changepoint.np defines no logLik method;

  • a change in mean under the default "MBIC" penalty. Loading changepoint.np (which this package imports, so it is always loaded) replaces changepoint's logLik method for cpt objects with one that errors on exactly that combination. Any other penalty ("BIC", "AIC", a numeric value) reports normally, as do change_in = "var" and "meanvar".