R help - Jun 2005 - A suggestion to improve ifelse behaviour with vector yes/noarguments

> Thomas Lumley wrote:
> > On Tue, 31 May 2005, Duncan Murdoch wrote:
> > 
> > 
> >>M??kinen Jussi wrote:
> >>
> >>>Dear All,
> >>>
> >>>I luckily found the following feature (or problem) when tried
to
> >>>apply
> >>>ifelse-function to an ordered data.
> >>>
> >>>
> >>>
> >>>>test <- c(TRUE, TRUE, TRUE, FALSE, FALSE, FALSE, FALSE)
ifelse(test,
> >>>>0, 1:4)
> >>>
> >>>[1] 0 0 0 4 1 2 3
> >>>
> > 
> > <snippage>
> > 
> >>As Dimitris said, this is just recycling.  I think getting rid of 
> >>recycling
> >>on vectors with length greater than 1 would have been a good
decision in S
> >>about 15 years ago, but it's too late now.
> > 
> > 
> > It wouldn't help the original poster, though.  I agree that 
> > 0,0,0,4,1,2,3
> > is a slightly weird result, but I can't think of any reasonable
model for
> > the behaviour of ifelse() that would give any other result except an
error
> > message. [or  0,NA,NA,4,NA,NA,NA, I suppose].
> 
> I would vote for the error message.  I can't think of a single example 
> where a vector of length 7 is needed, and a vector of length 4 is 
> recycled to give it, that *doesn't* give a slightly weird result.
> 
> Maybe this is something that should have been changed in R 2.0.0; we 
> squandered that change from 1.x.x to 2.x.x.
> 
> Duncan Murdoch
Hello,

I'm happy with the modified ifelse:

ifelse.o <- function (test, yes, no) 
{
    storage.mode(test) <- "logical"
    ans <- test
    nas <- is.na(test)
    if (any(test[!nas])) 
        ans[test & !nas] <- rep(yes, length.out = length(ans))[test &
            !nas]
    if (any(!test[!nas])) 
		### Changed
        ans[!test & !nas] <- rep(no, length.out = length(ans[!test &
!nas]))
    ans[nas] <- NA
    ans
}

giving me:
> test <- c(TRUE, TRUE, TRUE, FALSE, FALSE, FALSE, FALSE)
> ifelse.o(test, 0, 1:4)
 [1] 0 0 0 1 2 3 4

and
> test <- c(FALSE, FALSE, TRUE, TRUE, TRUE, FALSE, FALSE, FALSE, FALSE,
FALSE, FALSE)
> ifelse.o(test, 0, 1:4)
 [1] 1 2 0 0 0 3 4 1 2 3 4

comparing to:
> ifelse(test, 0, 1:4)
 [1] 1 2 0 0 0 2 3 4 1 2 3

So in 'ifelse.o' the recycling starts from the first element of the
'no' -argument and continues the cycling in a logical way (IMHO). That
is something I expected 'ifelse' would do.

Thank you for your comments,

Jussi M??kinen

M??kinen Jussi wrote:
> Hello,
> 
> I'm happy with the modified ifelse:
> 
> ifelse.o <- function (test, yes, no) 
> {
>     storage.mode(test) <- "logical"
>     ans <- test
>     nas <- is.na(test)
>     if (any(test[!nas])) 
>         ans[test & !nas] <- rep(yes, length.out = length(ans))[test
&
>             !nas]
>     if (any(!test[!nas])) 
> 		### Changed
>         ans[!test & !nas] <- rep(no, length.out = length(ans[!test
& !nas]))
>     ans[nas] <- NA
>     ans
> }
I wouldn't be:

 > x
  [1] -0.4539550 -1.3023478  0.9034912  1.7485065  0.6910265 -0.7712547
  [7] -0.6345585  1.8296632  2.1207810  0.7643834
 > ifelse.o(x > 0, 1:10, 0)
  [1]  0  0  3  4  5  0  0  8  9 10
 > ifelse.o(x > 0, 0, 1:10)
  [1] 1 2 0 0 0 3 4 0 0 0

I'd call these results fairly perverse.

Duncan Murdoch

Thanks Duncan,

For pointing this out. You maybe saved my day. I agree that these results are
neither something I would like to see. I changed the code to be:

ifelse.o <- function (test, yes, no) 
{
    storage.mode(test) <- "logical"
    ans <- test
    nas <- is.na(test)
    if (any(test[!nas])) 
        ### Own change
	  ans[test & !nas] <- rep(yes, length.out = length(ans[test &
!nas]))
    if (any(!test[!nas]))
	### Own change
        ans[!test & !nas] <- rep(no, length.out = length(ans[!test &
!nas]))
	ans[nas] <- NA
    ans
}

which yields now
> x <- rnorm(10)
> ifelse.o(x > 0, 1:10, 0)
 [1] 1 2 0 3 0 4 0 5 6 0
> ifelse.o(x > 0, 0, 1:10)
 [1] 0 0 1 0 2 0 3 0 0 4

I'm even more happy now :-) I will anyway need to use this modificate
function for just one specific problem. Hopefully there is no other hooks,

Jussi



M??kinen Jussi wrote:
> Hello,
> 
> I'm happy with the modified ifelse:
> 
> ifelse.o <- function (test, yes, no)
> {
>     storage.mode(test) <- "logical"
>     ans <- test
>     nas <- is.na(test)
>     if (any(test[!nas])) 
>         ans[test & !nas] <- rep(yes, length.out = length(ans))[test
&
>             !nas]
>     if (any(!test[!nas])) 
> 		### Changed
>         ans[!test & !nas] <- rep(no, length.out = length(ans[!test
& !nas]))
>     ans[nas] <- NA
>     ans
> }
I wouldn't be:

 > x
  [1] -0.4539550 -1.3023478  0.9034912  1.7485065  0.6910265 -0.7712547
  [7] -0.6345585  1.8296632  2.1207810  0.7643834
 > ifelse.o(x > 0, 1:10, 0)
  [1]  0  0  3  4  5  0  0  8  9 10
 > ifelse.o(x > 0, 0, 1:10)
  [1] 1 2 0 0 0 3 4 0 0 0

I'd call these results fairly perverse.

Duncan Murdoch