runsd {caTools} | R Documentation |
Moving (aka running, rolling) Window's Standard Deviation calculated over a vector
runsd(x, k, center = runmean(x,k), endrule=c("sd", "NA", "trim", "keep", "constant", "func"))
x |
numeric vector of length n |
k |
width of moving window; must be an integer between one and n. In case
of even k's one will have to provide different center function, since
runmed does not take even k's. |
endrule |
character string indicating how the values at the beginning
and the end, of the data, should be treated. Only first and last k2
values at both ends are affected, where k2 is the half-bandwidth
k2 = k %/% 2 .
endrule in runmed function which has the
following options: “c("median", "keep", "constant") ” .
|
center |
moving window center. Defaults
to running mean (runmean function). Similar to center
in mad function. |
Apart from the end values, the result of y = runmad(x, k) is the same as
“for(j=(1+k2):(n-k2)) y[j]=sd(x[(j-k2):(j+k2)], na.rm = TRUE)
”. It can handle
non-finite numbers like NaN's and Inf's (like mean(x, na.rm = TRUE)
).
The main incentive to write this set of functions was relative slowness of
majority of moving window functions available in R and its packages. With the
exception of runmed
, a running window median function, all
functions listed in "see also" section are slower than very inefficient
“apply(embed(x,k),1,FUN)
” approach.
Returns a numeric vector of the same length as x
. Only in case of
endrule="trim"
.the output will be shorter.
Jarek Tuszynski (SAIC) jaroslaw.w.tuszynski@saic.com
Links related to:
runsd
- sd
, rollVar
from
fSeries library
runmin
,
runmax
, runquantile
, runmad
and
runmean
apply
(embed(x,k), 1, FUN)
(fastest), rollFun
from fSeries (slow), running
from gtools
package (extremely slow for this purpose), rapply
from
zoo library, subsums
from
magic library can perform running window operations on data with any
dimensions.
# show runmed function k=25; n=200; x = rnorm(n,sd=30) + abs(seq(n)-n/4) col = c("black", "red", "green") m=runmean(x, k) y=runsd(x, k, center=m) plot(x, col=col[1], main = "Moving Window Analysis Functions") lines(m , col=col[2]) lines(m-y/2, col=col[3]) lines(m+y/2, col=col[3]) lab = c("data", "runmean", "runmean-runsd/2", "runmean+runsd/2") legend(0,0.9*n, lab, col=col, lty=1 ) # basic tests against apply/embed eps = .Machine$double.eps ^ 0.5 k=25 # odd size window a = runsd(x,k, endrule="trim") b = apply(embed(x,k), 1, sd) stopifnot(all(abs(a-b)<eps)); k=24 # even size window a = runsd(x,k, endrule="trim") b = apply(embed(x,k), 1, sd) stopifnot(all(abs(a-b)<eps)); # test against loop approach # this test works fine at the R prompt but fails during package check - need to investigate k=25; n=200; x = rnorm(n,sd=30) + abs(seq(n)-n/4) # create random data x[seq(1,n,11)] = NaN; # add NANs k2 = k k1 = k-k2-1 a = runsd(x, k) b = array(0,n) for(j in 1:n) { lo = max(1, j-k1) hi = min(n, j+k2) b[j] = sd(x[lo:hi], na.rm = TRUE) } #stopifnot(all(abs(a-b)<eps)); # compare calculation at array ends k=25; n=100; x = rnorm(n,sd=30) + abs(seq(n)-n/4) a = runsd(x, k, endrule="sd" ) # fast C code b = runsd(x, k, endrule="func") # slow R code stopifnot(all(abs(a-b)<eps)); # test if moving windows forward and backward gives the same results k=51; a = runsd(x , k) b = runsd(x[n:1], k) stopifnot(all(abs(a[n:1]-b)<eps)); # speed comparison ## Not run: x=runif(1e5); k=51; # reduce vector and window sizes system.time(runsd( x,k,endrule="trim")) system.time(apply(embed(x,k), 1, sd)) ## End(Not run)