vignettes/persephone-hierarchical.Rmd
persephone-hierarchical.RmdHierarchical time series are time series that can be divided into sub-series. In persephone, such objects are created similar to a nested list. The following example creates a series with two subseries a and b
library(persephone)
hts <- perHts(
a = perX13(AirPassengers),
b = perTramo(AirPassengers)
)
hts
#> component run class
#> FALSE hierarchicalTimeSeries
#> a FALSE x13Single
#> b FALSE tramoseatsSingleIf the run() method of hts is invoked, the following adjustments will be made
a+b is adjusted and the results are saved in hts$output.a is adjusted with x13()
b is adjusted with tramoseats()
The results can be viewed as follows
display_result <- function(ts) {
window(ts, end = c(1949, 6))
}
hts$run()
display_result(hts$adjusted)
#> Warning in (function () : The decision between direct and indirect adjustment was not recorded yet.
#> Direct adjustment is returned.
#> Jan Feb Mar Apr May Jun
#> 1949 245.3656 249.3203 250.2444 254.9434 254.7441 252.3840
display_result(hts$components$a$adjusted)
#> Jan Feb Mar Apr May Jun
#> 1949 123.7166 125.2532 125.9332 128.1540 129.0103 126.8570
display_result(hts$adjustedIndirect)
#> Jan Feb Mar Apr May Jun
#> 1949 246.3994 249.9133 251.0553 255.6256 256.3824 253.0490The last line displays the results of an indirect adjustment which amounts to a$adjusted + b$adjusted in this case.
In case more hierarchy levels are introduced, the calls to perHts can be nested.
hts_nested <- perHts(
hts = hts,
hts_child = perHts(
a = perX13(AirPassengers)
),
b = perTramo(AirPassengers)
)
hts_nested
#> component run class seasonality logTransform
#> FALSE hierarchicalTimeSeries <NA> NA
#> hts TRUE hierarchicalTimeSeries Present TRUE
#> hts/a TRUE x13Single Present TRUE
#> hts/b TRUE tramoseatsSingle Present TRUE
#> hts_child FALSE hierarchicalTimeSeries <NA> NA
#> hts_child/a FALSE x13Single <NA> NA
#> b FALSE tramoseatsSingle <NA> NA
#> arimaMdl nOutliers qStat
#> <NA> NA NA
#> (0 1 0)(0 1 1) 2 NA
#> (0 1 0)(0 1 1) 1 0.22
#> (0 1 0)(0 1 1) 2 NA
#> <NA> NA NA
#> <NA> NA NA
#> <NA> NA NAAs we can see, the runs of hts are available to this object.
To iterate over all components of a perHts object, the iterate method can be used. It baehaves similar to lapply.
hts_nested$run()
hts_nested$iterate(asTable = TRUE, function(obj) {
list(mean = mean(obj$adjusted),
var = sd(obj$adjusted))
})
#> component mean var
#> 1 1120.0985 444.2033
#> 2 hts 560.0492 222.1017
#> 3 hts/a 280.0419 111.5113
#> 4 hts/b 280.0246 111.0508
#> 5 hts_child 280.0246 111.0508
#> 6 hts_child/a 280.0419 111.5113
#> 7 b 280.0246 111.0508
hts_nested$iterate(asTable = TRUE, persephone:::generateQrList)[, 1:9]
#> component Method Period Nobs Start End Log.Transformation
#> 1 TS 12 144 1-1949 12-1960 TRUE
#> 2 hts TS 12 144 1-1949 12-1960 TRUE
#> 3 hts/a x13 12 144 1-1949 12-1960 TRUE
#> 4 hts/b TS 12 144 1-1949 12-1960 TRUE
#> 5 hts_child TS 12 144 1-1949 12-1960 TRUE
#> 6 hts_child/a x13 12 144 1-1949 12-1960 TRUE
#> 7 b TS 12 144 1-1949 12-1960 TRUE
#> ARIMA.Model LeapYear
#> 1 (0 1 0)(0 1 1) TRUE
#> 2 (0 1 0)(0 1 1) TRUE
#> 3 (0 1 0)(0 1 1) FALSE
#> 4 (0 1 0)(0 1 1) TRUE
#> 5 (0 1 0)(0 1 1) TRUE
#> 6 (0 1 0)(0 1 1) FALSE
#> 7 (0 1 0)(0 1 1) TRUE
hts_nested$setOptions(spec = "RSA1", recursive = TRUE)
get_spec <- function(x) {
list(spec = x$spec)
}
## check if options where applied correctly
hts_nested$iterate(asTable = TRUE, get_spec)
#> component spec
#> 1 RSA1
#> 2 hts RSA1
#> 3 hts/a RSA1
#> 4 hts/b RSA1
#> 5 hts_child RSA1
#> 6 hts_child/a RSA1
#> 7 b RSA1
## modify options beginning from hts
hts$setOptions(spec = "RSA0", recursive = TRUE)
hts_nested$iterate(asTable = TRUE, get_spec)
#> component spec
#> 1 RSA1
#> 2 hts RSA1
#> 3 hts/a RSA0
#> 4 hts/b RSA0
#> 5 hts_child RSA1
#> 6 hts_child/a RSA1
#> 7 b RSA1