changepoint.mv (version 1.0.2)

mrc: Most Recent Changepoints.

Description

Detects the Most Recent Changepoints (mrc) for panel data consisting of many related univariate timeseries (Bardwell, Eckley, Fearnhead and Smith, 2016). The method first determines the most recent univariate changepoints using PELT (Killick, Fearnhead and Eckley 2012) and then pools information across the time-series by solving the K-median problem using tb.raw (Teitz and Bart, 1968).

Usage

mrc(
  data,
  cost = "mean",
  alpha = function(n) 1.5 * log(n),
  pmax = 5,
  indexed = FALSE,
  mad = FALSE,
  phi = 0
)

Arguments

data

An $n\times N$ matrix or data frame representing a length $n$ time series containing observations of $N$ variables. The data can contain missing values which are indicated by NA.

cost

A string indicating which cost function to use. Possible choices are "mean" (change in mean) or "var" (change in variance). The default value is cost="mean".

alpha

The variable-specific penalty used to penalise the addition of a given changepoint into a given variable. This can be specified as a real positive value or as a function of $n$. The latter form is used when the data contains missing values which leads to time series of different lengths.

Default value alpha=function(n) 1.5*log(n).

pmax

Maximum number of most recent changepoints to search for. Default value pmax=5.

indexed

Boolean value indicating that the first column of data is an index variable. If indexed=FALSE an index variable will automatically be generated. Default value is indexed=FALSE.

mad

Boolean value indicating if the variates should be scaled by an estimate of the their standard deviation obtained using mean absolute deviation (Zhang, Nancy, Siegmund and David 2007). This is useful in conjunction with cost="mean" for which unit variance is assumed. Default value is mad=FALSE.

phi

Lag 1 autocorrelation to model the temporal dependence in the residuals of the time series assuming a MA(1) process. Default phi=0.

References

doi:10.1080/00401706.2018.1438926changepoint.mv

OR:TEITZBARTchangepoint.mv

doi:10.1080/01621459.2012.737745changepoint.mv

Examples

# NOT RUN {
library(changepoint.mv)
data(mrcexample)
res<-mrc(mrcexample,pmax=2)
MDL(res)     # MDL == pmax (possibly under-estimating MDL, retry)
res<-mrc(mrcexample,pmax=6)
MDL(res)     # MDL = 5 (< pmax)
# view the most recent changepoints (corresponding to pmax = 5)
unique(cpts.mr(res,p=5)[,1])
summary(res) # summary of result

# }