sapplyMpfr: Apply a Function over a "mpfr" Vector

In Rmpfr: Interface R to MPFR - Multiple Precision Floating-Point Reliable

Apply a Function over a "mpfr" Vector

Description

Users may be disappointed to note that sapply() or vapply() typically do not work with "mpfr" numbers.

This is a simple (but strong) approach to work around the problem, based on lapply().

Usage

sapplyMpfr(X, FUN, ..., drop_1_ = TRUE)

Arguments

X	a vector, possibly of class "mpfr".
FUN	a function returning an "mpfr" vector or even an "mpfrArray". May also be a function returning a numeric vector or array for numeric X, and which returns "mpfr(Array)" for an X argument inheriting from "mpfr".
...	further arguments passed to lapply, typically further arguments to FUN.
drop_1_	logical (with unusual name on purpose!) indicating if 1-column matrices ("mpfrMatrix") should be “dropped” to vectors ("mpfr"), the same as in base R's own sapply. This has been implicitly FALSE in Rmpfr versions 0.8-5 to 0.8-9 (Oct 2021 to June 2022), accidentally. Since Rmpfr 0.9-0, this has been made an argument with default TRUE to be compatible by default with R's sapply.

Details

In the case FUN(<length-1>) returns an array or "mpfrArray", i.e., with two or more dimensions, sapplyMpfr() returns an "mpfrArray"; this is analogous to sapply(X, FUN, simplify = "array") (rather than the default sapply() behaviour which returns a matrix also when a higher array would be more “logical”.)

Value

an "mpfr" vector, typically of the same length as X.

Note

This may still not always work as well as sapply() does for atomic vectors. The examples seem to indicate that it typically does work as desired, since Rmpfr version 0.9-0.

If you want to transform back to regular numbers anyway, it maybe simpler and more efficient to use

    res <- lapply(....)
    sapply(res, asNumeric, simplify = "array")
  

instead of sapplyMpfr().

Author(s)

Martin Maechler

See Also

sapply, lapply, etc.

Examples

sapplyMpfr0 <- ## Originally, the function was simply defined as
  function (X, FUN, ...) new("mpfr", unlist(lapply(X, FUN, ...), recursive = FALSE))

(m1 <- sapply    (     3,      function(k) (1:3)^k)) # 3 x 1  matrix (numeric)
(p1 <- sapplyMpfr(mpfr(3, 64), function(k) (1:3)^k))
stopifnot(m1 == p1, is(p1, "mpfrMatrix"), dim(p1) == c(3,1), dim(p1) == dim(m1))
k.s <- c(2, 5, 10, 20)
(mk <- sapply    (     k.s,      function(k) (1:3)^k)) # 3 x 4    "       "
(pm <- sapplyMpfr(mpfr(k.s, 64), function(k) (1:3)^k))
stopifnot(mk == pm, is(pm, "mpfrMatrix"), dim(pm) == 3:4, 3:4 == dim(mk))
## was *wrongly* 4x3  in Rmpfr 0.8-x
f5k  <- function(k) outer(1:5, k+0:2, `^`)# matrix-valued
(mk5 <- sapply    (     k.s,      f5k))  # sapply()'s default; not "ideal"
(ak5 <- sapply    (     k.s,      f5k, simplify = "array")) # what we want
(pm5 <- sapplyMpfr(mpfr(k.s, 64), f5k))
stopifnot(c(mk5) == c(ak5), ak5 == pm5, is(pm5, "mpfrArray"), is.array(ak5),
          dim(pm5) == dim(ak5), dim(pm5) == c(5,3, 4))
if(require("Bessel")) { # here X, is simple
  bI1 <- function(k) besselI.nuAsym(mpfr(1.31e9, 128), 10, expon.scaled=TRUE, k.max=k)
  bImp1 <- sapplyMpfr (0:4, bI1, drop_1_ = FALSE) # 1x5 mpfrMatrix -- as in DPQ 0.8-8
  bImp  <- sapplyMpfr (0:4, bI1, drop_1_ = TRUE ) # 5 "mpfr" vector {by default}
  bImp0 <- sapplyMpfr0(0:4, bI1) # 5-vector
  stopifnot(identical(bImp, bImp0), bImp == bImp1,
            is(bImp, "mpfr"), is(bImp1, "mpfrMatrix"), dim(bImp1) == c(1, 5))
}# {Bessel}

Rmpfr documentation built on Nov. 18, 2024, 3:01 p.m.