aggregate.zoo: Compute Summary Statistics of zoo Objects

In parsifal9/test: S3 Infrastructure for Regular and Irregular Time Series (Z's Ordered Observations)

Description

Splits a "zoo" object into subsets along a coarser index grid, computes summary statistics for each, and returns the reduced "zoo" object.

Usage

## S3 method for class 'zoo'
aggregate(x, by, FUN = sum, ...,
  regular = NULL, frequency = NULL, coredata = TRUE)

Arguments

x	an object of class "zoo".
by	index vector of the same length as index(x) which defines aggregation groups and the new index to be associated with each group. If by is a function, then it is applied to index(x) to obtain the aggregation groups.
FUN	a function to compute the summary statistics which can be applied to all subsets. Always needs to return a result of fixed length (typically scalar).
...	further arguments passed to FUN.
regular	logical. Should the aggregated series be coerced to class "zooreg" (if the series is regular)? The default is FALSE for "zoo" series and TRUE for "zooreg" series.
frequency	numeric indicating the frequency of the aggregated series (if a "zooreg" series should be returned. The default is to determine the frequency from the data if regular is TRUE. If frequency is specified, it sets regular to TRUE. See examples for illustration.
coredata	logical. Should only the coredata(x) be passed to every by group? If set to FALSE the full zoo series is used.

Value

An object of class "zoo" or "zooreg".

Note

The xts package functions endpoints, period.apply to.period, to.weekly, to.monthly, etc., can also directly input and output certain zoo objects and so can be used for aggregation tasks in some cases as well.

See Also

zoo

Examples

## averaging over values in a month:
# x.date is jan 1,3,5,7; feb 9,11,13; mar 15,17,19
x.date <- as.Date(paste(2004, rep(1:4, 4:1), seq(1,20,2), sep = "-")); x.date
x <- zoo(rnorm(12), x.date); x
# coarser dates - jan 1 (4 times), feb 1 (3 times), mar 1 (3 times)
x.date2 <- as.Date(paste(2004, rep(1:4, 4:1), 1, sep = "-")); x.date2
x2 <- aggregate(x, x.date2, mean); x2
# same - uses as.yearmon
x2a <- aggregate(x, as.Date(as.yearmon(time(x))), mean); x2a
# same - uses by function
x2b <- aggregate(x, function(tt) as.Date(as.yearmon(tt)), mean); x2b
# same - uses cut
x2c <- aggregate(x, as.Date(cut(time(x), "month")), mean); x2c
# almost same but times of x2d have yearmon class rather than Date class
x2d <- aggregate(x, as.yearmon, mean); x2d

# compare time series
plot(x)
lines(x2, col = 2)

## aggregate a daily time series to a quarterly series
# create zoo series
tt <- as.Date("2000-1-1") + 0:300
z.day <- zoo(0:300, tt)

# function which returns corresponding first "Date" of quarter
first.of.quarter <- function(tt) as.Date(as.yearqtr(tt))

# average z over quarters
# 1. via "yearqtr" index (regular)
# 2. via "Date" index (not regular)
z.qtr1 <- aggregate(z.day, as.yearqtr, mean)
z.qtr2 <- aggregate(z.day, first.of.quarter, mean)

# The last one used the first day of the quarter but suppose
# we want the first day of the quarter that exists in the series
# (and the series does not necessarily start on the first day
# of the quarter).
z.day[!duplicated(as.yearqtr(time(z.day)))]

# This is the same except it uses the last day of the quarter.
# It requires R 2.6.0 which introduced the fromLast= argument.
## Not run: 
z.day[!duplicated(as.yearqtr(time(z.day)), fromLast = TRUE)]

## End(Not run)

# The aggregated series above are of class "zoo" (because z.day
# was "zoo"). To create a regular series of class "zooreg",
# the frequency can be automatically chosen
zr.qtr1 <- aggregate(z.day, as.yearqtr, mean, regular = TRUE)
# or specified explicitely
zr.qtr2 <- aggregate(z.day, as.yearqtr, mean, frequency = 4)


## aggregate on month and extend to monthly time series
if(require(chron)) {
y <- zoo(matrix(11:15, nrow = 5, ncol = 2), chron(c(15, 20, 80, 100, 110)))
colnames(y) <- c("A", "B")

# aggregate by month using first of month as times for coarser series
# using first day of month as repesentative time
y2 <- aggregate(y, as.Date(as.yearmon(time(y))), head, 1)

# fill in missing months by merging with an empty series containing
# a complete set of 1st of the months
yrt2 <- range(time(y2))
y0 <- zoo(,seq(from = yrt2[1], to = yrt2[2], by = "month"))
merge(y2, y0)
}

# given daily series keep only first point in each month at
# day 21 or more
z <- zoo(101:200, as.Date("2000-01-01") + seq(0, length = 100, by = 2))
zz <- z[as.numeric(format(time(z), "%d")) >= 21]
zz[!duplicated(as.yearmon(time(zz)))]

# same except times are of "yearmon" class
aggregate(zz, as.yearmon, head, 1)

# aggregate POSIXct seconds data every 10 minutes
Sys.setenv(TZ = "GMT")
tt <- seq(10, 2000, 10)
x <- zoo(tt, structure(tt, class = c("POSIXt", "POSIXct")))
aggregate(x, time(x) - as.numeric(time(x)) %% 600, mean)

# aggregate weekly series to a series with frequency of 52 per year
suppressWarnings(RNGversion("3.5.0"))
set.seed(1)
z <- zooreg(1:100 + rnorm(100), start = as.Date("2001-01-01"), deltat = 7)

# new.freq() converts dates to a grid of freq points per year
# yd is sequence of dates of firsts of years
# yy is years of the same sequence
# last line interpolates so dates, d, are transformed to year + frac of year
# so first week of 2001 is 2001.0, second week is 2001 + 1/52, third week
# is 2001 + 2/52, etc.
new.freq <- function(d, freq = 52) {
       y <- as.Date(cut(range(d), "years")) + c(0, 367)
       yd <- seq(y[1], y[2], "year")
       yy <- as.numeric(format(yd, "%Y"))
       floor(freq * approx(yd, yy, xout = d)$y) / freq
}

# take last point in each period
aggregate(z, new.freq, tail, 1)

# or, take mean of all points in each
aggregate(z, new.freq, mean)

# example of taking means in the presence of NAs
z.na <- zooreg(c(1:364, NA), start = as.Date("2001-01-01"))
aggregate(z.na, as.yearqtr, mean, na.rm = TRUE)

# Find the sd of all days that lie in any Jan, all days that lie in
# any Feb, ..., all days that lie in any Dec (i.e. output is vector with
# 12 components)
aggregate(z, format(time(z), "%m"), sd)

parsifal9/test documentation built on Dec. 31, 2020, 1:14 a.m.