Mix test signal

A piecewise-constant/linear signal from the literature.

Usage

signal_mix(n = 2000, seed = NULL)

Arguments

n

Length of the signal. Defaults to 2000.

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.

Value

A tibble with columns index and value.

See also

Other test signals: cpt_simulate(), signal_blocks(), signal_fms(), signal_stairs(), signal_teeth()

Examples

x <- signal_mix(seed = 2026)
attr(x, "true_changepoints")
#> [1]  300  700  900 1300 1600
# the linear ramps are not level shifts, so a mean-change detector puts
# changepoints inside them rather than at the segment joins
cpt_detect(x$value, method = "pelt")$changepoints$cp
#> Warning: `pelt` is scale-sensitive for a change in mean (its cost assumes unit noise), and this series' noise standard deviation is about 0.48. On data ten times larger it over-segments badly; on data ten times smaller it finds nothing. Standardise the series first (e.g. `x / 0.48`), or use `change_in = "meanvar"`, which estimates the noise per segment.
#> [1]  406  569 1012 1138 1257 1705 1893