seqXtend function

Sequence Covering the Range of X, including X

Produce a sequence of unique values (sorted increasingly), containing the initial set of values x. This can be useful for setting prediction e.g. ranges in nonparametric regression.

seqXtend(x, length., method = c("simple", "aim", "interpolate"),
        from = NULL, to = NULL)

Arguments

• x: numeric vector.

• length.: integer specifying approximately the desired length() of the result.

• method: string specifying the method to be used. The default, "simple" uses seq(*, length.out = length.) where "aim" aims a bit better towards the desired final length, and "interpolate" interpolates evenly inside

  each interval $(x[i], x[i+1])$ in a way to make all the new intervalls of approximately the same length.

• from, to: numbers to be passed to (the default method for) seq(), defaulting to the minimal and maximal x

  value, respectively.

Note

method = "interpolate" typically gives the best results. Calling roundfixS, it also need more computational resources than the other methods.

Returns

numeric vector of increasing values, of approximate length length.

(unless length. < length(unique(x)) in which case, the result is simply sort(unique(x))), containing the original values of x.

From, r <- seqXtend(x, *), the original values are at indices ix <- match(x,r), i.e., identical(x, r[ix]).

Author(s)

Martin Maechler

See Also

seq; plotDS can make particularly good use of seqXtend()

Examples

a <- c(1,2,10,12)
seqXtend(a, 12)# --> simply 1:12
seqXtend(a, 12, "interp")# ditto
seqXtend(a, 12, "aim")# really worse
stopifnot(all.equal(seqXtend(a, 12, "interp"), 1:12))

## for a "general" x, however, "aim" aims better than default
x <- c(1.2, 2.4, 4.6, 9.9)
length(print(seqXtend(x, 12)))        # 14
length(print(seqXtend(x, 12, "aim"))) # 12
length(print(seqXtend(x, 12, "int"))) # 12

## "interpolate" is really nice:
xt <- seqXtend(x, 100, "interp")
plot(xt, main="seqXtend(*, 100, \"interpol\")")
points(match(x,xt), x, col = 2, pch = 20)
# .... you don't even see that it's not equidistant
# whereas the cheap method shows ...
xt2 <- seqXtend(x, 100)
plot(xt2, col="blue")
points(match(x,xt2), x, col = 2, pch = 20)

## with "Date" objects
Drng <- as.Date(c("2007-11-10", "2012-07-12"))
(px <- pretty(Drng, n = 16)) # say, for the main labels
## say, a finer grid, for ticks -- should be almost equidistant
n3 <- 3*length(px)
summary(as.numeric(diff(seqXtend(px, n3))))        # wildly varying
summary(as.numeric(diff(seqXtend(px, n3, "aim")))) #   (ditto)
summary(as.numeric(diff(seqXtend(px, n3, "int")))) # around 30