This vignette introduces the DTSg package, shows how to create objects of its main as well as only class and explains their two interfaces: R6 as code and S3 in comments. Familiarity with the data.table package helps better understanding certain parts of the vignette, but is not essential to follow it.
First, let's load some data. The package is shipped with a data.table containing a daily time series of river flows:
library(data.table)
library(DTSg)
data(flow)
flow
#> date flow
#> 1: 2007-01-01 9.540
#> 2: 2007-01-02 9.285
#> 3: 2007-01-03 8.940
#> 4: 2007-01-04 8.745
#> 5: 2007-01-05 8.490
#> ---
#> 2165: 2012-12-27 26.685
#> 2166: 2012-12-28 28.050
#> 2167: 2012-12-29 23.580
#> 2168: 2012-12-30 18.840
#> 2169: 2012-12-31 17.250
summary(flow)
#> date flow
#> Min. :2007-01-01 00:00:00 Min. : 4.995
#> 1st Qu.:2008-07-19 00:00:00 1st Qu.: 8.085
#> Median :2010-01-12 00:00:00 Median : 11.325
#> Mean :2010-01-08 23:32:46 Mean : 16.197
#> 3rd Qu.:2011-07-08 00:00:00 3rd Qu.: 18.375
#> Max. :2012-12-31 00:00:00 Max. :290.715
Now that we have a data set, we can create our first object by providing it to the new() method of the package's main R6 class generator DTSg. In addition, we specify an ID in order to give the object a name:
TS <- DTSg$new(
values = flow,
ID = "River Flow"
)
Creating an object with the package's alternative interface abusing an S4 constructor looks like this:
TS <- new(
Class = "DTSg",
values = flow,
ID = "River Flow"
)
Printing the object shows us, among others, the data provided, the specified ID and that the object represents a regular UTC time series with a periodicity of one day and 2192 timestamps. It also shows us that the first column has been renamed to .dateTime. This column serves as the object's time index and cannot be changed at will:
TS$print() # or 'print(TS)' or just 'TS'
#> Values:
#> .dateTime flow
#> <POSc> <num>
#> 1: 2007-01-01 9.540
#> 2: 2007-01-02 9.285
#> 3: 2007-01-03 8.940
#> 4: 2007-01-04 8.745
#> 5: 2007-01-05 8.490
#> ---
#> 2188: 2012-12-27 26.685
#> 2189: 2012-12-28 28.050
#> 2190: 2012-12-29 23.580
#> 2191: 2012-12-30 18.840
#> 2192: 2012-12-31 17.250
#>
#> ID: River Flow
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 2192
With this done, we can move on and further explore our newly created time series data object with a summary (summary()), a report on missing values (nas()) and a plot (plot()). It suddenly seems to contain several missing values, which apparently were not there upon loading the data set (plot() requires the dygraphs and RColorBrewer packages to be installed; HTML vignettes unfortunately cannot display interactive elements, hence I included a static image of the JavaScript chart instead):
TS$summary() # or 'summary(TS)'
#> flow
#> Min. : 4.995
#> 1st Qu.: 8.085
#> Median : 11.325
#> Mean : 16.197
#> 3rd Qu.: 18.375
#> Max. :290.715
#> NA's :23
TS$nas(cols = "flow") # or 'nas(TS, cols = "flow")'
#> .col .group .from .to .n
#> 1: flow 1 2007-10-12 2007-10-24 13
#> 2: flow 2 2007-10-26 2007-11-03 9
#> 3: flow 3 2007-11-10 2007-11-10 1
if (requireNamespace("dygraphs", quietly = TRUE) &&
requireNamespace("RColorBrewer", quietly = TRUE)) {
TS$plot(cols = "flow") # or 'plot(TS, cols = "flow")'
}
Looking at the original data set reveals that the missing values implicitly already were there. Putting the data set into the object simply expanded it to the automatically detected periodicity and made them explicit (this behaviour can be changed, however, DTSg objects work best with explicitly missing values):
flow[date >= as.POSIXct("2007-10-09", tz = "UTC") & date <= as.POSIXct("2007-11-13", tz = "UTC"), ]
#> date flow
#> 1: 2007-10-09 9.180
#> 2: 2007-10-10 9.075
#> 3: 2007-10-11 9.000
#> 4: 2007-10-25 18.150
#> 5: 2007-11-04 25.350
#> 6: 2007-11-05 23.550
#> 7: 2007-11-06 23.400
#> 8: 2007-11-07 26.400
#> 9: 2007-11-08 39.150
#> 10: 2007-11-09 37.200
#> 11: 2007-11-11 27.450
#> 12: 2007-11-12 39.750
#> 13: 2007-11-13 31.350
For fairly small gaps like this it might be okay to fill them by means of linear interpolation. Using the colapply() method together with the interpolateLinear() function will do the trick:
TS <- TS$colapply(fun = interpolateLinear)
# or 'colapply(TS, fun = interpolateLinear)'
TS$nas()
#> Empty data.table (0 rows and 5 cols): .col,.group,.from,.to,.n
In case no column name is provided through the cols argument of the colapply() method, the first numeric column is taken by default. Column names for the cols argument can be requested from DTSg objects with the help of the cols() method. It, among others, supports a class and/or pattern argument:
TS$cols() # or 'cols(TS)'
#> [1] "flow"
TS$cols(class = "numeric") # or 'cols(TS, class = "numeric")'
#> [1] "flow"
TS$cols(class = "character")
#> character(0)
TS$cols(class = c("double", "integer")) # class of column flow is numeric
#> character(0)
TS$cols(pattern = "f.*w") # or 'cols(TS, pattern = "f.*w")'
#> [1] "flow"
TS$cols(pattern = "temp")
#> character(0)
The time series reaches from the start of the year 2007 to the end of the year 2012. Let's say we are only interested in the first two years. With alter() we can shorten, lengthen and/or change the periodicity of a DTSg object. The latter can be achieved through its by argument (no example given):
TS <- TS$alter(from = "2007-01-01", to = "2008-12-31")
# or 'alter(TS, from = "2007-01-01", to = "2008-12-31")'
TS
#> Values:
#> .dateTime flow
#> <POSc> <num>
#> 1: 2007-01-01 9.540
#> 2: 2007-01-02 9.285
#> 3: 2007-01-03 8.940
#> 4: 2007-01-04 8.745
#> 5: 2007-01-05 8.490
#> ---
#> 727: 2008-12-27 18.180
#> 728: 2008-12-28 16.575
#> 729: 2008-12-29 13.695
#> 730: 2008-12-30 12.540
#> 731: 2008-12-31 11.940
#>
#> ID: River Flow
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 731
In order to get mean monthly river flows as an example, we can use the aggregate() method with one of the package's temporal aggregation level functions (TALFs) as its funby argument (the fun argument also accepts a named character vector or list of functions, which allows for calculating several summary statistics at once):
TSm <- TS$aggregate(funby = byYm____, fun = mean)
# or 'aggregate(TS, funby = byYm____, fun = mean)'
TSm
#> Values:
#> .dateTime flow
#> <POSc> <num>
#> 1: 2007-01-01 25.281290
#> 2: 2007-02-01 14.496964
#> 3: 2007-03-01 12.889839
#> 4: 2007-04-01 12.470500
#> 5: 2007-05-01 9.233226
#> ---
#> 20: 2008-08-01 12.641129
#> 21: 2008-09-01 13.710500
#> 22: 2008-10-01 10.626774
#> 23: 2008-11-01 8.902000
#> 24: 2008-12-01 16.435645
#>
#> ID: River Flow
#> Aggregated: TRUE
#> Regular: FALSE
#> Periodicity: 1 months
#> Min lag: Time difference of 28 days
#> Max lag: Time difference of 31 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 24
Printing the aggregated object shows us that its aggregated field has been set to TRUE. This is merely an indicator telling us to now interpret the timestamps of the series as periods between subsequent timestamps and not as snap-shots anymore.
The one family of temporal aggregation level functions of the package sets timestamps to the lowest possible point in time of the corresponding temporal aggregation level, i.e. truncates timestamps, and the other family extracts a certain part of them. An example is given for quarters below. By convention, the year is set to 2199 in the latter case:
TSQ <- TS$aggregate(funby = by_Q____, fun = mean)
# or 'aggregate(TS, funby = by_Q____, fun = mean)'
TSQ
#> Values:
#> .dateTime flow
#> <POSc> <num>
#> 1: 2199-01-01 18.46127
#> 2: 2199-04-01 12.95266
#> 3: 2199-07-01 13.48924
#> 4: 2199-10-01 15.84620
#>
#> ID: River Flow
#> Aggregated: TRUE
#> Regular: FALSE
#> Periodicity: 3 months
#> Min lag: Time difference of 90 days
#> Max lag: Time difference of 92 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 4
Additional temporal aggregation level functions exist for years, days, hours, minutes and seconds. Furthermore, the temporal aggregation level of certain temporal aggregation level functions can be adjusted via the multiplier argument of the aggregate() as well as other methods, e.g. multiplier = 6 makes byYm____ aggregate to half years instead of months.
The last thing we want to achieve for now is the calculation of moving averages for a window of two time steps before and after each timestamp. We can do so with the help of the rollapply() method:
TSs <- TS$rollapply(fun = mean, na.rm = TRUE, before = 2, after = 2)
# or 'rollapply(TS, fun = mean, na.rm = TRUE, before = 2, after = 2)'
TSs
#> Values:
#> .dateTime flow
#> <POSc> <num>
#> 1: 2007-01-01 9.2550
#> 2: 2007-01-02 9.1275
#> 3: 2007-01-03 9.0000
#> 4: 2007-01-04 8.7720
#> 5: 2007-01-05 8.5710
#> ---
#> 727: 2008-12-27 19.3080
#> 728: 2008-12-28 16.4520
#> 729: 2008-12-29 14.5860
#> 730: 2008-12-30 13.6875
#> 731: 2008-12-31 12.7250
#>
#> ID: River Flow
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 731
On a side note, some of the methods, which take a function as an argument (colapply() and rollapply()) hand over to it an additional list argument called .helpers containing useful data for the development of user defined functions (please see the respective help pages for further information). This can of course be a problem for functions like cumsum(), which do not expect such a thing. A solution to this is to set the helpers argument of the respective method to FALSE.
With this said, let's join the result of the last calculation to the original time series data object and extract its values as a data.table for further processing in a final step (please note that the .dateTime column gets its original name back):
TS <- TS$merge(y = TSs, suffixes = c("_orig", "_movavg"))
# or 'merge(TS, y = TSs, suffixes = c("_orig", "_movavg"))'
TS$values()
#> date flow_orig flow_movavg
#> 1: 2007-01-01 9.540 9.2550
#> 2: 2007-01-02 9.285 9.1275
#> 3: 2007-01-03 8.940 9.0000
#> 4: 2007-01-04 8.745 8.7720
#> 5: 2007-01-05 8.490 8.5710
#> ---
#> 727: 2008-12-27 18.180 19.3080
#> 728: 2008-12-28 16.575 16.4520
#> 729: 2008-12-29 13.695 14.5860
#> 730: 2008-12-30 12.540 13.6875
#> 731: 2008-12-31 11.940 12.7250
Better ways to add results of the colapply() or rollapply() method as new column(s) are actually their resultCols and suffix arguments. Their use would not have allowed for demonstrating the merge() method though.
For a full explanation of all the methods and functions available in the package as well as their arguments please consult the help pages. Especially the following methods and functions have not been discussed in this vignette:
getCol() or `[.DTSg`()refresh()rowaggregate()rowbind()setColNames()setCols()subset()rollback()S3WrapperGenerator()The fields of a DTSg object of which the metadata are part of can be accessed through so called active bindings:
TS$ID
#> [1] "River Flow"
Valid results are returned for the following fields:
aggregatedfastfunbyApproachIDna.statusparameterperiodicityregulartimestampstimezoneunitvariantA subset of this fields can also be actively set (please note that reference semantics always apply to fields, hence the “largely” in the title of the package):
# two new `DTSg` objects in order to demonstrate reference semantics, which are
# propagated by assignments and broken by deep clones
TSassigned <- TS
TScloned <- TS$clone(deep = TRUE) # or 'clone(x = TS, deep = TRUE)'
# set new ID
TS$ID <- "Two River Flows"
TS
#> Values:
#> .dateTime flow_orig flow_movavg
#> <POSc> <num> <num>
#> 1: 2007-01-01 9.540 9.2550
#> 2: 2007-01-02 9.285 9.1275
#> 3: 2007-01-03 8.940 9.0000
#> 4: 2007-01-04 8.745 8.7720
#> 5: 2007-01-05 8.490 8.5710
#> ---
#> 727: 2008-12-27 18.180 19.3080
#> 728: 2008-12-28 16.575 16.4520
#> 729: 2008-12-29 13.695 14.5860
#> 730: 2008-12-30 12.540 13.6875
#> 731: 2008-12-31 11.940 12.7250
#>
#> ID: Two River Flows
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 731
# due to reference semantics, the new ID is also propagated to `TSassigned`, but
# not to `TScloned` (as all data manipulating methods create a deep clone by
# default, it is usually best to set or update fields after and not before
# calling such a method)
TSassigned
#> Values:
#> .dateTime flow_orig flow_movavg
#> <POSc> <num> <num>
#> 1: 2007-01-01 9.540 9.2550
#> 2: 2007-01-02 9.285 9.1275
#> 3: 2007-01-03 8.940 9.0000
#> 4: 2007-01-04 8.745 8.7720
#> 5: 2007-01-05 8.490 8.5710
#> ---
#> 727: 2008-12-27 18.180 19.3080
#> 728: 2008-12-28 16.575 16.4520
#> 729: 2008-12-29 13.695 14.5860
#> 730: 2008-12-30 12.540 13.6875
#> 731: 2008-12-31 11.940 12.7250
#>
#> ID: Two River Flows
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 731
TScloned
#> Values:
#> .dateTime flow_orig flow_movavg
#> <POSc> <num> <num>
#> 1: 2007-01-01 9.540 9.2550
#> 2: 2007-01-02 9.285 9.1275
#> 3: 2007-01-03 8.940 9.0000
#> 4: 2007-01-04 8.745 8.7720
#> 5: 2007-01-05 8.490 8.5710
#> ---
#> 727: 2008-12-27 18.180 19.3080
#> 728: 2008-12-28 16.575 16.4520
#> 729: 2008-12-29 13.695 14.5860
#> 730: 2008-12-30 12.540 13.6875
#> 731: 2008-12-31 11.940 12.7250
#>
#> ID: River Flow
#> Aggregated: FALSE
#> Regular: TRUE
#> Periodicity: Time difference of 1 days
#> Missing values: explicit
#> Time zone: UTC
#> Timestamps: 731
The fields, which cannot be actively set are:
regulartimestampsPlease refer to the help pages for further details.