On Fri, 20 Jan 2006, Peter Dalgaard wrote:
[...]
> Yes, this is a bit nasty, but... What is happening is similar to this:
>
>> d <- data.frame(a=factor(LETTERS), b=factor(letters))
>> d[1,]
> a b
> 1 A a
>> as.character(d[1,])
> [1] "1" "1"
>> as.character(d[1,1])
> [1] "A"
>> as.character(d[1,1,drop=F])
> [1] "1"
>
> or this:
>
>> l <- list(a=factor("x"),b=factor("y"))
>> l
> $a
> [1] x
> Levels: x
>
> $b
> [1] y
> Levels: y
>
>> as.character(l)
> [1] "1" "1"
>
> The thing is that as.character on a list will first coerce factors to
> numeric, then numeric to character.
Nope. It just coerces an INTSXP to a STRSXP. as.character (and all other
forms of coercion that I can think of quickly) ignores classes except when
initially dispatching.
Note that these examples are special cases:
> as.character(d[1:2,])
[1] "c(1, 2)" "c(1, 2)"
may also be unexpected but follows from the general (undocumented, I
dare say) rules.
> I'm not sure whether there could be a rationale for it, but it isn't
> S-PLUS compatible (not 6.2.1 anyway, which is the most recent one that I
> have access to).
My S-PLUS deparses:
> l <- list(a=factor("x"),b=factor("y"))
> as.character(l)
[1] "structure(.Data = 1, .Label = \"x\", class = \"factor\")"
[2] "structure(.Data = 1, .Label = \"y\", class = \"factor\")"
which seems no better (and probably worse).
The only other consistent option I can see is for all coercion methods to
dispatch at each element of a recursive object, which I suspect introduces
a considerable overhead for very little gain.
One could perhaps argue for a data.frame method, since coercion operations
on dataframes are rare and that is a case where people get factors where
they wanted character columns.
--
Brian D. Ripley, [EMAIL PROTECTED]
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
