Documentation
documentation.RmdThe documentation covers the three parts of every detection: the cost function, the segmentation method and model selection. This page lists the basic functions and shows how the package is organised; the sidebar links to each part.
Basic functions
Every segmentation class (binSeg, Window,
PELT and Dynp) has the same methods, so code
written for one runs with the others.
| Function | What it does |
|---|---|
costFunc$new() |
Chooses the cost function, which sets the kind of change to look for. |
binSeg$new(), Window$new(),
PELT$new(), Dynp$new()
|
Chooses the segmentation method, which searches for the change-points. |
$fit(tsMat) |
Attaches the time series: a matrix with one row per time point. |
$predict(pen = ...) |
Returns the change-points for a penalty per change-point. |
$predict(nBkps = ...) |
Returns the change-points for a given number of change-points. |
$plot() |
Plots the series with the detected segments. |
$segments() |
Gives the cost and parameter estimates of each segment. |
$eval(a, b) |
Gives the cost of the segment (a, b]. |
How the package is organised
A detection with K change-points t_1 < \dots < t_K, where t_0 = 0 and t_{K+1} = n, minimises the total cost of the segments plus a penalty \lambda per change-point:
\sum_{k=0}^{K} c\left(y_{(t_k+1):t_{k+1}}\right) + \lambda K.
The cost function sets c, the segmentation method searches over the change-points, and model selection sets \lambda, or fixes K instead. The package follows the same split.
rupturesRcpp-
Cost functions
What kind of change to look for
-
costFunc$new():"L1","L2","SIGMA","VAR","LinearL2","LinearSIGMA","LinearL1","Custom" -
costFactory: costs without a detection
-
-
Segmentation methods
How to search for change-points
-
Model selection
How many change-points to keep
-
$predict(pen = ...): a penalty per change-point -
$predict(nBkps = ...): a fixed number -
$getHistory()and$plotElbow(): the elbow method
-