onlineCPD (version 1.0)

onlineCPD: Online Changepoint Detection

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.

Value

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

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.

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)