aggregateSeries
Time Series and Signal Aggregation

Description

Aggregation and coursening of time series and signals. This is the method for the aggregate function for objects of class timeSeries and signalSeries.

Usage

aggregateSeries(x, pos, FUN, moving=FALSE, together=FALSE, drop.empty=TRUE,
                include.ends=FALSE, adj, offset, colnames, by, k.by=1,
                week.align=NULL, holidays=timeDate(), align.by=TRUE,
                incr=1, ...)

Arguments

x the series object to be aggregated.
pos the break positions for aggregation (can also be supplied in the by argument if desired).
FUN the function to use for aggregation. Often mean, sum, or hloc. If moving is FALSE, FUN can also be a character string like "fastFFF", to aggregate column-wise using the corresponding fast igroupFFF function. (The together argument is ignored.)

Possible choices for FFF are currently Means, Maxs, Mins, Sums, Prods, Anys, and Alls.

moving either FALSE to do standard aggregation, or a positive integer N to perform a moving aggregation (normally used for a moving average) over N samples.
together a logical value. if TRUE, passes all columns of x together into FUN; If FALSE (the default), passes each column separately into FUN for each aggregation block.
drop.empty a logical value. If TRUE (the default), drops aggregation blocks with no positions to aggregate from the output.
include.ends a logical value. If TRUE, includes the positions before the first given aggregation block and after the last in the first/last blocks. If FALSE (the default), does not include those positions in the output.
adj if provided, adjusts the positions of the output series so that they lie a fraction adj towards the blocks ending position. The default is to use the lower end of the block for the output position. 0.5 uses the center of the aggregation block for the output position, and 1.0 uses the upper end of the block.
offset as an alternative to adj, provide a constant offset to add to the lower end of the aggregation block to get the output series positions. For instance, in monthly aggregation, you might supply an offset of 9 days so that the output positions fell on the tenth of each month.
colnames new column names for the output series. The default is to use the same names as the input series if the output series has the same width.
by if pos is missing and by is supplied for aggregating calendar-based time series, new positions are generated as a regular time/date sequence using by, k.by, week.align, and holidays.

by gives the spacing between successive values in the sequence. This can be a timeSpan, timeRelative, or numeric value, in which case k.by is ignored.

Alternatively, it can be one of the following character strings:

  • "milliseconds"
  • "seconds"
  • "minutes"
  • "hours"
  • "days"
  • "weekdays"
  • "bizdays"
  • "weeks"
  • "months"
  • "quarters"
  • "years"
giving the time units of intervals between values in the sequence.
k.by a non-zero integer giving the width of the interval between consecutive values in the sequence in terms of the units given in by. Ignored if by is not a character string or if pos is given.
week.align if not NULL and by is "weeks", you can supply a character string (or a number, 0 to 6 with 0 being Sunday) to specify a weekday to use for aggregation. The character string must be sufficient to make a unique case-insensitive match to the strings in timeSeriesOptions("time.day.name"). Ignored if by is not a character string or pos is given.
holidays holidays for business day sequences. (Ignored if by is not a character string or pos is given.)
align.by a logical value. If TRUE (the default), adjustd the sequence so that each element is on a whole number of the by * k.by units. For example, if the units are 2 months, make the sequence be only on the first of January, March, and so on. Ignored if by is not a character string.
incr For moving aggregation, the moving window moves forward by this many steps in the time series, for each window.
... Additional arguments to pass to FUN.
Value
returns a new time series whose positions are the adjusted passed-in positions or positions generated from by, k.by, and so on, (or possibly a subset if drop.empty is TRUE) and whose rows are aggregated from x as specified in the arguments. Aggregation takes place by separating x into blocks separated by the positions (or overlapping blocks with a fixed number of samples if moving is supplied), and then applying FUN to each column (or all columns together if together is TRUE) and forming a new time series with the positions and aggregated data.
See Also
timeSeries, signalSeries, align, aggregate
Examples
x <- timeSeries(data.frame(1:20,rnorm(20)), timeCalendar(d=1:20))
aggregate(x, FUN=mean, by="weeks")
Package splusTimeSeries version 6.1.1-7
Package Index