README.md

In lpishchagina/FPOPdim2: FPOPdim2

FPOPdim2 Vignette

Liudmila Pishchagina

march 1, 2021

Quick Start

FPOPdim2 is an R package written in Rcpp/C++ and developed to detection changepoints using the Functional Pruning Optimal Partitioning method (FPOP) in bivariate time series of length n.

The package implements the following types of pruning:

type = 0: PELT method;

type = 1: ("intersection" of sets), approximation - "rectangle";

type = 2:("intersection" of sets)"minus"("union" of sets), approximation - "rectangle";

type = 3: (last disk)"minus"("union" of sets), approximation - "disk";

type = 5: ...;

We present a basic use of the main functions of the FPOPdim2 package.

We install the package from Github:

#devtools::install_github("lpishchagina/FPOPdim2")
library(FPOPdim2)

The function data_gen2D

The data_gen2D function simulates a bivariate time series of length n with the arguments:

n is the time series length.

chpts is the changepoint vector that gives the last index of each segment.

The last element of chpts is always less than to the length of time series (by default, chpts = NULL for the data without changepoints).

means1 vector of successive means for the first univariate time series (by default it is equal to 0).

means2 vector of successive means for the second univariate time series (by default it is equal to 0).

noise is a variance of the time series (by default it is equal to 1).

set.seed(13)

size <- 1000

Data1 <- data_gen2D(size) 

Data2 <-  data_gen2D(n = size, chpts = 50, means1 = c(0,1), means2 = c(0,1))

The function FPOP2D

The FPOP2D function returns the result of the segmentation of FPOP-method using the parameters:

data1 is the first time series.

data2 is the second time series.

penalty is the value of penalty (a non-negative real number).

The penalty here equals to a classic 2*(noise^2)*log(n).

type is a value defining the type of geometry for FPOP-pruning.

The type must be either 0, 1, 2, 3 or 5:

type = 0: PELT method;

type = 1: ("intersection" of sets), approximation - "rectangle";

type = 2:("intersection" of sets)"minus"("union" of sets), approximation - "rectangle";

type = 3: (last disk)"minus"("union" of sets), approximation - "disk";

type = 5: .....

We choose a Gaussian cost in 2-dimension.

library(base)

library(FPOPdim2)

set.seed(13)

size <- 1000

beta <- 4 * log(size)

Data1 <- data_gen2D(size) 

FPOP2D(Data1[1,], Data1[2,], penalty = beta, type = 0)

FPOP2D(Data1[1,], Data1[2,], penalty = beta, type = 1)

FPOP2D(Data1[1,], Data1[2,], penalty = beta, type = 2)

FPOP2D(Data1[1,], Data1[2,], penalty = beta, type = 3)

FPOP2D(Data1[1,], Data1[2,], penalty = beta, type = 5)

$chpts
numeric(0)

$means1
[1] -0.003112923

$means2
[1] 0.0008128666

$globalCost
[1] 1978.728

library(base)

library(FPOPdim2)

set.seed(13)

size <- 1000

beta <- 4 * log(size)

Data2 <- data_gen2D(n = size, chpts = 50, means1 = c(0,1), means2 = c(0,1))

FPOP2D(Data2[1,], Data2[2,], penalty = beta, type = 0)

FPOP2D(Data2[1,], Data2[2,], penalty = beta, type = 1)

FPOP2D(Data2[1,], Data2[2,], penalty = beta, type = 2)

FPOP2D(Data2[1,], Data2[2,], penalty = beta, type = 3)

FPOP2D(Data2[1,], Data2[2,], penalty = beta, type = 5)

$chpts
[1] 49

$means1
[1] -0.05569976  0.99854507

$means2
[1] -0.08093126  1.00397318

$globalCost
[1] 1974.816

chpts is the changepoint vector that gives the last index of each segment.

For the data without changepoints chpts = numeric(0).

means1 is the vector of successive means for the first time series.

means2 is the vector of successive means for the second time series.

globalCost is the overall Gaussian cost of the segmented data.

Back to Top

lpishchagina/FPOPdim2 documentation built on Dec. 21, 2021, 11:50 a.m.