sorting variable names containing digits

mixedsort in gtools will give the same result as mysort(s) but
differs in the case of t.

On Sun, Dec 21, 2008 at 8:33 PM, John Fox <jfox at mcmaster.ca> wrote:

Dear r-helpers,

I'm looking for a way of sorting variable names in a "natural"

order, when

the names are composed of digits and other characters. I know that

this is a

vague idea, and that sorting character strings is a complex topic,

but

perhaps a couple of examples will clarify what I mean:

s <- c("x1b", "x1a", "x02b", "x02a", "x02", "y1a1", "y10a2",

+   "y10a10", "y10a1", "y2", "var10a2", "var2", "y10")

sort(s)

 [1] "var10a2" "var2"    "x02"     "x02a"    "x02b"    "x1a"
 [7] "x1b"     "y10"     "y10a1"   "y10a10"  "y10a2"   "y1a1"
[13] "y2"

mysort(s)

 [1] "var2"    "var10a2" "x1a"     "x1b"     "x02"     "x02a"
 [7] "x02b"    "y1a1"    "y2"      "y10"     "y10a1"   "y10a2"
[13] "y10a10"

t <- c("q10.1.1", "q10.2.1", "q2.1.1", "q10.10.2")

sort(t)

[1] "q10.1.1"  "q10.10.2" "q10.2.1"  "q2.1.1"

mysort(t)

[1] "q2.1.1"   "q10.1.1"  "q10.2.1"  "q10.10.2"

Here, sort() is the standard R function and mysort() is a

replacement, which

sorts the names into the order that seems natural to me, at least

in the

cases that I've tried:

mysort <- function(x){
 sort.helper <- function(x){
   prefix <- strsplit(x, "[0-9]")
   prefix <- sapply(prefix, "[", 1)
   prefix[is.na(prefix)] <- ""
   suffix <- strsplit(x, "[^0-9]")
   suffix <- as.numeric(sapply(suffix, "[", 2))
   suffix[is.na(suffix)] <- -Inf
   remainder <- sub("[^0-9]+", "", x)
   remainder <- sub("[0-9]+", "", remainder)
   if (all (remainder == "")) list(prefix, suffix)
   else c(list(prefix, suffix), Recall(remainder))
   }
 ord <- do.call("order", sort.helper(x))
 x[ord]
  }

I have a couple of applications in mind, one of which is

recognizing

repeated-measures variables in "wide" longitudinal datasets, which

often are

named in the form x1, x2, ... , xn.

mysort(), which works by recursively slicing off pairs of non-digit

and

digit strings, seems more complicated than it should have to be,

and I

wonder whether anyone has a more elegant solution. I don't think

that

efficiency is a serious issue for the applications I'm considering,

but of

course a more efficient solution would be of interest.

Thanks,
 John

------------------------------
John Fox, Professor
Department of Sociology
McMaster University
Hamilton, Ontario, Canada
web: socserv.mcmaster.ca/jfox