ts_detect: Change point detection

In connorbrubaker/rulsif.ts: Time Series Change Point Detection Using Relative Unconstrained Least Squares Importance Fitting (RULSIF)

Description

Detect change points in a time series using RelULSIF.

Usage

ts_detect(
  ts,
  window_size = 5,
  step = NULL,
  alpha = 0.05,
  k = 100,
  n_folds = 5,
  thresh = 0.9,
  make_plot = FALSE
)

Arguments

ts	Time series to detect change points in. Assumes this is a D by N matrix where D is the dimension of the time series and N is the number of time points. If a vector is provided, will assume the time series is one-dimensional.
window_size	The length of the sub-sequences generated from the series. Default 5.
step	How many sub-sequences forward and backward to from a time point to compute a score from. Default is 10% of the length of the series if not specified.
alpha	Relative parameter in [0, 1). Default 0.05. Setting to 0 recovers ordinary unconstrained least squares importance fitting.
k	Number of basis functions. Default is minimum of 100 and dimension of the time series.
n_folds	Number of folds to use in determining optimal kernel bandwidth and lambda parameter in RULSIF.
thresh	Scalar in (0, 1) indicating the percentile above which a score is considered a potential change-point. Lower values increase the sensitivity.
make_plot	Logical. On the same figure, make a plot of each dimension of the time series, the rPE scores, and highlight in the time series plots the change points detected in red. Default FALSE.

Value

List of 3:

• step: the step used

• scores: rPE scores

• change_points: time points that a change was detected at the given threshold

Examples

s <- c(rnorm(150, mean = 0), rnorm(150, mean = 5), rnorm(150, mean = 1))
s <- matrix(s, nrow = 1)
ts_detect(s)

connorbrubaker/rulsif.ts documentation built on Dec. 19, 2021, 6:02 p.m.