g13ndf detects change points in a univariate time series, that is, the time points at which some feature of the data, for example the mean, changes. Change points are detected using binary segmentation using one of a provided set of cost functions.
Let denote a series of data and denote a set of ordered (strictly monotonic increasing) indices known as change points, with and . For ease of notation we also define . The change points, , split the data into segments, with the th segment being of length and containing .
Given a cost function, , g13ndf gives an approximate solution to
where is a penalty term used to control the number of change points. The solution is obtained in an iterative manner as follows:
Set , and
Set . If , where is a user-supplied control parameter, then terminate the process for this segment.
Find that minimizes
If inequality (1) is false then the process is terminated for this segment.
If inequality (1) is true, then is added to the set of change points, and the segment is split into two subsegments, and . The whole process is repeated from step 2 independently on each subsegment, with the relevant changes to the definition of and (i.e., is set to when processing the left hand subsegment and is set to when processing the right hand subsegment.
The change points are ordered to give .
g13ndf supplies four families of cost function. Each cost function assumes that the series, , comes from some distribution, . The parameter space, is subdivided into containing those parameters allowed to differ in each segment and those parameters treated as constant across all segments. All four cost functions can then be described in terms of the likelihood function, and are given by:
where the is the maximum likelihood estimate of within the th segment. Four distributions are available; Normal, Gamma, Exponential and Poisson distributions. Letting
the log-likelihoods and cost functions for the four distributions, and the available subdivisions of the parameter space are:
Both mean and variance change:
when calculating for the Poisson distribution, the sum is calculated for rather than .
Chen J and Gupta A K (2010) Parametric Statistical Change Point Analysis With Applications to GeneticsMedicine and FinanceSecond Edition Birkhäuser
West D H D (1979) Updating mean and variance estimates: An improved method Comm. ACM22 532–555
1: – IntegerInput
On entry: a flag indicating the assumed distribution of the data and the type of change point being looked for.
Data from a Normal distribution, looking for changes in the mean, .
Data from a Normal distribution, looking for changes in the standard deviation .
Data from a Normal distribution, looking for changes in the mean, and standard deviation .
Data from a Gamma distribution, looking for changes in the scale parameter .
Data from an exponential distribution, looking for changes in .
Data from a Poisson distribution, looking for changes in .
, , , , or .
2: – IntegerInput
On entry: , the length of the time series.
3: – Real (Kind=nag_wp) arrayInput
On entry: , the time series.
if , that is the data is assumed to come from a Poisson distribution, is used in all calculations.
if , or , , for ;
if , each value of y must be representable as an integer;
if , each value of y must be small enough such that, for , can be calculated without incurring overflow.
4: – Real (Kind=nag_wp)Input
On entry: , the penalty term.
There are a number of standard ways of setting , including:
SIC or BIC
where is the number of parameters being treated as estimated in each segment. This is usually set to when and otherwise.
If no penalty is required then set . Generally, the smaller the value of the larger the number of suggested change points.
5: – IntegerInput
On entry: the minimum distance between two change points, that is .
6: – IntegerInput
On entry: if distributional parameters have been supplied in param.
if , ;
otherwise or .
7: – Real (Kind=nag_wp) arrayInput
On entry: , values for the parameters that will be treated as fixed. If then param is not referenced.
if , , the standard deviation of the Normal distribution, is estimated from the full input data. Otherwise .
If , , the mean of the Normal distribution, is estimated from the full input data. Otherwise .
If , must hold the shape, , for the Gamma distribution, otherwise param is not referenced.
if or , .
8: – IntegerInput
On entry: , the maximum depth for the iterative process, which in turn puts an upper limit on the number of change points with .
If then no limit is put on the depth of the iterative process and no upper limit is put on the number of change points, other than that inherent in the length of the series and the value of minss.
9: – IntegerOutput
On exit: , the number of change points detected.
10: – Integer arrayOutput
Note: the dimension of the array tau
must be at least
if , and at least otherwise.
On exit: the first elements of tau hold the location of the change points. The th segment is defined by to , where and .
On entry: ifail must be set to , . If you are unfamiliar with this argument you should refer to Section 3.4 in How to Use the NAG Library and its Documentation for details.
For environments where it might be inappropriate to halt program execution when an error is detected, the value is recommended. If the output of error messages is undesirable, then the value is recommended. Otherwise, if you are not familiar with this argument, the recommended value is . When the value is used it is essential to test the value of ifail on exit.
On exit: unless the routine detects an error or a warning has been flagged (see Section 6).
Error Indicators and Warnings
If on entry or , explanatory error messages are output on the current error message unit (as defined by x04aaf).
Errors or warnings detected by the routine:
On entry, . Constraint: , , , , or .
On entry, . Constraint: .
On entry, and . Constraint: if , or then , for .
On entry, , is too large.
On entry, . Constraint: .
On entry, . Constraint: if then or .
On entry, . Constraint: if then .
On entry, and . Constraint: if or and , then .
To avoid overflow some truncation occurred when calculating the cost function, . All output is returned as normal.
To avoid overflow some truncation occurred when calculating the parameter estimates returned in sparam. All output is returned as normal.
An unexpected error has been triggered by this routine. Please
See Section 3.9 in How to Use the NAG Library and its Documentation for further information.
Your licence key may have expired or may not have been installed correctly.
See Section 3.8 in How to Use the NAG Library and its Documentation for further information.
Dynamic memory allocation failed.
See Section 3.7 in How to Use the NAG Library and its Documentation for further information.
The calculation of means and sums of squares about the mean during the evaluation of the cost functions are based on the one pass algorithm of West (1979) and are believed to be stable.
Parallelism and Performance
g13ndf is threaded by NAG for parallel execution in multithreaded implementations of the NAG Library.
Please consult the X06 Chapter Introduction for information on how to control and interrogate the OpenMP environment used within this routine. Please also consult the Users' Note for your implementation for any additional implementation-specific information.
This example identifies changes in the mean, under the assumption that the data is normally distributed, for a simulated dataset with observations. A BIC penalty is used, that is , the minimum segment size is set to and the variance is fixed at across the whole input series.