This section will run the basic onlineCPD function to output an ocpd object from the simulated data.
Printing the results shows a list of the attributes returned, and the main finding which is the output list of changepoints.
Following is some explanation of each attribute returned.
Main Outputs
The main output from the OCPD objects are the changepoints list and the R matrix if the it is chosen to be returned.
- Changepoints List: list of changepoints computed from different approaches
- colmaxes: the list of changepoints taken from the maximum probability in each column of the R matrix
- threshchps: the list of changepoints that have run length probability above a threshold value, e.g. 0.5
- maxCPs: the list of changepoints with the highest overall probability
The list of possible changepoints can be accessed from the attribute “changepoint_lists”. The code and output from this example is shown below.
cpdf<- data.frame(method=names(ocpd1$changepoint_lists))
cpdf$changepoints<- unlist(ocpd1$changepoint_lists, recursive = FALSE)
kable(cpdf)
colmaxes |
c(1, 51, 71) |
threshcps |
c(1, 51, 71) |
maxCPs |
c(1, 51, 71) |
Additional Outputs
Additional outputs are also returned - these are needed for running the algorithm online since this algorithm can work being called point by point to build on a pre-existing ocpd object, but not needed for interpreting the final results. These outputs are:
- ocpd_settings: stores the settings used in running the ocpd function
- prevR: the column of the R matrix for the previous step
- prevRprod, prevRsum: versions of the R matrix containing cumulative product or sum along diagonals
- prevDataPt: the previous data point processed - needed for replacing missing data options
- time: the total number of points processed within this ocpd object
- update_paramsT: the current parameters used in the UPM update functions
- update_params0: the initial parameters used in the UPM update functions
- init_params: the params used to initialize the UMP update_params0
- logprobcps, logprobmaxes: the list of changepoints with the highest probability at each step, and the probability associated respectively
- currmu: the current mu compute after each timepoint