There are a number of changepoint detection packages in R. These packages have various APIs and sometimes it is difficult to compare the changepoints detected by various packages on the same data. Moreover, visualizing these changepoints along with the raw data is not connected to ggplot2. What the ggchangepoint package does is it combines a few functions from several changepoint packages together and offers the ggplot2 style changepoint plots.
In this introduction, all the data sets used are the same as Killick and Eckley (2014) starting from the Section 4.1, and all the variables are defined the same as the reference. Instead of following the authors’ code, we use ggchangepoint to carry out all the analysis, and users can make reasonable comparisons.
library(ggchangepoint)
library(ggplot2)
theme_set(theme_light())
set.seed(10)
<- c(rnorm(100, 0, 1),
m.data rnorm(100, 1, 1),
rnorm(100, 0, 1),
rnorm(100, 0.2, 1))
Using PELT
as the changepoint method:
<- cpt_wrapper(m.data, change_in = "mean", cp_method = "PELT")
m.pelt
m.pelt #> # A tibble: 2 x 2
#> cp cp_value
#> <int> <dbl>
#> 1 97 -1.61
#> 2 192 2.00
ggcptplot(m.data, change_in = "mean", cp_method = "PELT") +
labs(title = "Changes in mean (PELT)")
Using BinSeg
as the changepoint method:
<- cpt_wrapper(m.data, change_in = "mean", cp_method = "BinSeg")
m.binseg
m.binseg#> # A tibble: 2 x 2
#> cp cp_value
#> <dbl> <dbl>
#> 1 79 -1.05
#> 2 192 2.00
ggcptplot(m.data, change_in = "mean", cp_method = "BinSeg") +
labs(title = "Changes in mean (BinSeg)")
PELT
with penalty = "Manual"
:
<- cpt_wrapper(m.data,
m.pm change_in = "mean",
penalty = "Manual",
pen.value = "1.5 * log(n)")
m.pm#> # A tibble: 3 x 2
#> cp cp_value
#> <int> <dbl>
#> 1 97 -1.61
#> 2 192 2.00
#> 3 273 -0.481
ggcptplot(m.data,
change_in = "mean",
penalty = "Manual",
pen.value = "1.5 * log(n)",
cptline_color = "red",
cptline_size = 2) +
labs(title = "Changes in mean (PELT with Manual penalty)")
BinSeg
with penalty = "Manual"
:
<- cpt_wrapper(m.data,
m.bsm change_in = "mean",
cp_method = "BinSeg",
penalty = "Manual",
pen.value = "1.5 * log(n)")
m.bsm#> # A tibble: 4 x 2
#> cp cp_value
#> <dbl> <dbl>
#> 1 79 -1.05
#> 2 99 0.148
#> 3 192 2.00
#> 4 273 -0.481
ggcptplot(m.data,
change_in = "mean",
cp_method = "BinSeg",
penalty = "Manual",
pen.value = "1.5 * log(n)",
cptline_size = 2) +
labs(title = "Changes in mean (BinSeg with Manual penalty)")
Since Killick and Eckley (2014) used a data set from Lai et al. (2005) about genomic hy- bridization (aCGH). Here we also use the data set:
data("Lai2005fig4", package = "changepoint")
cpt_wrapper(Lai2005fig4$GBM29, change_in = "mean")
#> # A tibble: 6 x 2
#> cp cp_value
#> <int> <dbl>
#> 1 81 0.0330
#> 2 85 4.72
#> 3 89 0.299
#> 4 96 3.99
#> 5 123 1.23
#> 6 133 4.47
And we can quickly visualize it:
ggcptplot(Lai2005fig4$GBM29, change_in = "mean") +
labs(title = "GBM29 Changes in mean (PELT)")
We can also use ecp_wrapper()
to carry out the similar task on the data:
set.seed(2022)
ecp_wrapper(Lai2005fig4$GBM29)
#> # A tibble: 8 x 2
#> cp cp_value
#> <dbl> <dbl>
#> 1 29 1.31
#> 2 33 0.728
#> 3 82 4.51
#> 4 86 0.718
#> 5 90 5.62
#> 6 97 0.698
#> 7 124 4.59
#> 8 134 0.0302
set.seed(2022)
ggecpplot(Lai2005fig4$GBM29) +
labs(title = "GBM29 Changes using the ecp package")
We can use ggcptplot()
to detect and visualize changepoints detected by various functions. Otherwise, various packages need to be used.
Killick and Eckley (2014) referenced the Irish wind speeds data set, which has previously been analyzed by Haslett and Raftery (1989) and Gneiting, Genton, and Guttorp (2006). It is a data set from the gstat package.
data("wind", package = "gstat")
<- cpt_wrapper(diff(wind[, 11]), change_in = "var", cp_method = "BinSeg")
wind.bs
wind.bs#> # A tibble: 1 x 2
#> cp cp_value
#> <dbl> <dbl>
#> 1 2971 13.2
ggcptplot(diff(wind[, 11]), change_in = "var", cp_method = "BinSeg")
There is only one changepoint detected in the data.
Since cpt.meanvar()
is the default setting of cpt_wrapper()
. There will be no extra specification in the change_in
argument.
data("discoveries", package = "datasets")
<- cpt_wrapper(discoveries, test.stat = "Poisson")
dis.pelt
dis.pelt#> # A tibble: 3 x 2
#> cp cp_value
#> <int> <dbl>
#> 1 24 3
#> 2 29 9
#> 3 73 3
ggcptplot(discoveries, test.stat = "Poisson",
cptline_color = "red",
cptline_size = 2) +
ggtitle("The Discoveries dataset with identified changepoints (PELT)")
ggcptplot(discoveries, test.stat = "Poisson",
cp_method = "BinSeg",
cptline_color = "red",
cptline_size = 2) +
ggtitle("The Discoveries dataset with identified changepoints (BinSeg)")