Installation
You can install the released version of ChangePointTaylor from CRAN with:
Example
Load the package and other needed libraries for this example
library(ChangePointTaylor)
library(dplyr)
#>
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#>
#> filter, lag
#> The following objects are masked from 'package:base':
#>
#> intersect, setdiff, setequal, union
library(ggplot2)View the example dataset of US trade deficit data from January 1987 to December 1988.
US_Trade_Deficit
#> # A tibble: 24 x 2
#> date deficit_billions
#> <chr> <dbl>
#> 1 Jan '87 10.7
#> 2 Feb '87 13
#> 3 Mar '87 11.4
#> 4 Apr '87 11.5
#> 5 May '87 12.5
#> 6 Jun '87 14.1
#> 7 Jul '87 14.8
#> 8 Aug '87 14.1
#> 9 Sep '87 12.6
#> 10 Oct '87 16
#> # ... with 14 more rowsPlot the data
trade_deficit_plot <- US_Trade_Deficit %>%
mutate(date = as.Date(paste(date, "1"), format = "%b '%y %d")) %>%
ggplot(aes(x = date, y = deficit_billions, group = 1)) +
geom_line() +
geom_point() +
theme_bw() +
scale_x_date(date_breaks = "1 month", date_labels = "%b '%y") +
theme(
axis.text.x = element_text(angle = 45, vjust = 1, hjust =1),
axis.title.x = element_blank()
) +
ggtitle("US Trade Deficit: 1987-1988")
trade_deficit_plot
In its simplest form, the change_point_analyzer() function simply takes a numeric vector and returns the identified change points. However, the output only identifies changes by their index in the original numeric vector.
change_point_analyzer(US_Trade_Deficit$deficit_billions)
#> 2 Change(s) Identified
#> NA supplied to 'label' argument
#> # A tibble: 2 x 6
#> change_ix label `CI (95%)` change_conf From To
#> <dbl> <lgl> <chr> <dbl> <dbl> <dbl>
#> 1 6 NA (5 - 7) 0.917 11.8 14.3
#> 2 11 NA (11 - 11) 0.998 14.3 10.2When a vector of labels, the same length as the x values, is supplied to the label argument, those labels will be displayed in the output dataframe.
change_points <- change_point_analyzer(US_Trade_Deficit$deficit_billions, label = US_Trade_Deficit$date)
#> 2 Change(s) Identified
change_points
#> change_ix label CI (95%) change_conf From To
#> 1 6 Jun '87 (May '87 - Jul '87) 0.908 11.82 14.32
#> 2 11 Nov '87 (Nov '87 - Nov '87) 0.997 14.32 10.20Plot the change points we identified.
trade_deficit_plot +
geom_vline(xintercept = as.Date(paste(change_points$label, "1"), format = "%b '%y %d"), color = "steelblue", linetype = "dashed", size = 1.3)
The number of bootstraps can be controlled with the n_bootstraps argument. This can reduce stochastic differences between subsequent function calls; however, this comes at the expense of execution speed.
bench::mark(
change_point_analyzer(US_Trade_Deficit$deficit_billions, label = US_Trade_Deficit$date, n_bootstraps = 1000)
,change_point_analyzer(US_Trade_Deficit$deficit_billions, label = US_Trade_Deficit$date, n_bootstraps = 10000)
,check = F
,min_iterations = 2
,max_iterations = 5
) %>%
mutate(expression = c("1000 Bootstraps", "10000 Bootstraps")) %>%
select(expression:mem_alloc)
#> # A tibble: 2 x 5
#> expression min median `itr/sec` mem_alloc
#> <chr> <bch:tm> <bch:tm> <dbl> <bch:byt>
#> 1 1000 Bootstraps 232ms 247ms 4.10 134MB
#> 2 10000 Bootstraps 631ms 637ms 1.57 274MBThe the user can also adjust the minimum level of confidence a change point must reach to become an initial candidate (min_candidate_conf) and the minimum confidence to be included in the final table of change points (min_tbl_conf).