onlineCPD: Online Changepoint Detection

In onlineCPD: Detect Changepoints in Multivariate Time Series

Description

Performs the oCPD algorithm in a truely on-line fashion; each iteration, you provide it with one more data point in the series, and the function will update the run length probabilities.

Usage

onlineCPD(oCPD = NULL, datapt, timept = NULL, hazard_func = const_hazard)

Arguments

oCPD	an object of type "oCPD", datapt will be added to the end of this series
datapt	the next data point in the series. This must have the same number of variables as oCPD
timept	optional, the corresponding time for datapt
hazard_func	hazard function used in the model. Defaults to a constant hazard, suitable for exponential family models.

Details

This function performs the oCPD algorithm, one point at a time. To use, start with either the result to a call to offlineCPD or just a single data point. In the case of a single data point, leave argument oCPD as NULL. See Examples section.

See offlineCPD for more information on how the algorithm works.

Value

An object of class "oCPD", which is a list containing the following:

R	n by n matrix of run-length probabilities. The value at R[i,j] is the probability that at data point j, the current run length is i.
data	same as the input parameter, included for plotting.
time	same as the input parameter, included for plotting.
alpha	the vector of values of alpha after the final data point.
beta	the vector (or matrix) of values of beta (the variance) after the final data point.
kappa	the vector of values of kappa after the final data point.
mu	the vector (or matrix) of values of mu (the mean) after the final data point
max	vector of values; max[i] is the runlength with the highest probability. Used to plot the red diagonal line in plot.oCPD
changes	locations of detected changepoints. When the algorithm can not detect the exact location of a change, multiple possible values are reported.

Note

For a version of this same algorithm that functions offline (i.e. you give it the whole time series at once), see offlineCPD

This version runs a little slower because it needs to copy the "oCPD" object with every iteration. Use offlineCPD unless you don't have the whole time series available at the start.

Author(s)

Zachary Zanussi

References

Adams, R. P. and Mackay, D. J. C. (2007), Bayesian Online Changepoint Detection

####OUR PAPER, WHEN IT EXISTS

Examples

##### Univariate Data #####
set.seed(6)
x <- c(rnorm(50,mean=0.3,sd=0.15),rnorm(40,mean=0.7,sd=0.1),rnorm(60,mean=0.5,sd=0.15))
res <- onlineCPD(datapt=x[1])
for(k in x)
  res <- onlineCPD(res,k)
plot(res)

##### Real Multivariate Data #####
data(WalBelSentiment)
data(WalBelTimes)
res <- onlineCPD(datapt=WalBelSentiment[1400,],timept=WalBelTimes[1400])
for(k in 1401:1600)
  res <- onlineCPD(res,WalBelSentiment[k,],WalBelTimes[k])
plot(res)

## You can use onlineCPD to add points to an existing "oCPD" object
y <- c(rnorm(50,0.5,0.1),rnorm(20,0.48,0.02),rnorm(50,0.5,0.1))
res <- offlineCPD(y)
plot(res)
x <- rnorm(75,0.7,0.4)
for(k in x)
  res <- onlineCPD(res,k)
plot(res)

onlineCPD documentation built on Jan. 15, 2017, 7:22 p.m.