sorry, for joining this thread so late. I've just upgraded to R2.8.0
and was hit with the same issue immediately in my fts package.
Has any consensus been reached about how to deal with this issue?
Do I understand correctly that the change to use:
[sort.c: 592]
PROTECT(call = lang4(install(".gt"), x, si, sj));
c = asInteger(eval(call, rho));
inside of "static int equal(int i, int j, SEXP x, Rboolean nalast,
SEXP rho)" was to promote class based object dispatching for
comparisons?
-Whit
On Thu, Oct 23, 2008 at 6:16 AM, Gabor Grothendieck
<ggrothendieck at gmail.com> wrote:
Based on the quote that Peter gave:
o New generic function xtfrm() as an auxiliary helper for
sort(), order() and rank(). This should return a numeric
vector that sorts in the same way as its input. The default
method supports any class with ==, > and is.na() methods but
specific methods can be much faster.
As a side-effect, rank() will now work better on classed
objects, although possibly rather slowly.
it seems that it was the intention to have rank(x) use x only via xtfrm(x)
but that is not, in fact, how it works as defining xtfrm.zoo still
results in the
discussed error.
On Thu, Oct 23, 2008 at 5:13 AM, Pfaff, Bernhard Dr.
<Bernhard_Pfaff at fra.invesco.com> wrote:
Dear Achim, Gabor, Jeff, Peter and remaining list-subscribers, first, please accept my apologies for having started this thread. Given the avalanche of responses (some off-list) and the associated time stamps, some of you have almost pulled an all-nighter about this topic. I might have not have started this thread in the first place, if I would have known this. I agree with Achim's view that a "not working for zoo objects" strategy is preferable compared to a "half-working for zoo objects" strategy. I do not have either a problem by employing coredata(z) when necessary. Now, Gabor, you pointed out nicely that the culprit, namely that order() resides on top of xtfrm whereas rank() does not (this was voiced by you in an email to Peter and R-Devel, hence I inlucde these recipients again in this thread and the reason for doing this is also motivated by the following proposition): You further pointed out that the problems wrt to rank and zoo-objects could be solved if rank() would, like order() does, reside on top of xtfrm. My question/proposal would then be to follow this approach, i.e. use xtfrm in rank. Now, I am not that deep into R nor an expert to judge whether this would cause problems/breaking existing R code in other instances; hence I appreciate feedback if this would be a feasible/desirable change in R-Devel. Best, Bernhard
-----Urspr?ngliche Nachricht----- Von: Gabor Grothendieck [mailto:ggrothendieck at gmail.com] Gesendet: Donnerstag, 23. Oktober 2008 02:36 An: Peter Dalgaard Cc: Pfaff, Bernhard Dr.; r-devel at stat.math.ethz.ch Betreff: Re: [Rd] R 2.8.0 qqnorm produces error with object of class zoo? I don't think its hopeless. order works ok provided the underlying class defines an xtfrm method. I think rank should follow that route too. Its arguably the inconsistency between rank and order (order but not the rank uses xtfrm) that causes the inconsistent behavior between the two. If rank were also built on top of xtfrm then it would work as desired as well. On Wed, Oct 22, 2008 at 5:03 PM, Peter Dalgaard <p.dalgaard at biostat.ku.dk> wrote:
Gabor Grothendieck wrote:
And one other point. z <- zoo(1:4) .gt(z, 1, 2) fails because z[1] and z[2] are at different time points so z[1] == z[2] is logical(0) because when zoo compares objects it aligns them first.
Yes, that was the point that I was trying to make. Well,
arguably it doesn't
"fail", it just does what it is supposed to do. Things would
"work" with [[
or a preceding unclass(z), but that would break comparisons
involving, say,
POSIXlt objects. So you're sort of stuck between a rock and
a hard place.
On Wed, Oct 22, 2008 at 2:19 PM, Gabor Grothendieck <ggrothendieck at gmail.com> wrote:
Yes, I noticed that but rank is not generic. An xtfrm.zoo method has been added to zoo on R-Forge but rank still fails:
R.version.string
[1] "R version 2.8.0 Patched (2008-10-21 r46766)"
packageDescription("zoo")$Version
[1] "1.5-3"
library(zoo) # next line adds xtfrm zoo method xtfrm.zoo <- coredata z <- zoo(1:4) order(z) # ok
[1] 1 2 3 4
qqnorm(z) # ok rank(z) # error
Error in if (xi == xj) 0L else if (xi > xj) 1L else -1L : argument is of length zero
(If the MIME type is wrong, then that will happen.) Anyways, the root cause seems to be the new function
.gt() which is
related to
o New generic function xtfrm() as an auxiliary helper for
sort(), order() and rank(). This should return a numeric
vector that sorts in the same way as its input.
The default
method supports any class with ==, > and is.na()
methods but
specific methods can be much faster.
As a side-effect, rank() will now work better on classed
objects, although possibly rather slowly.
Here, "better" may be in the eyes of the beholder, for
dax[3]==dax[6]
Data: logical(0) Index: integer(0) and accordingly
rank(dax)
Error in if (xi == xj) 0L else if (xi > xj) 1L else -1L : argument is of length zero which is the error that you are seeing. What to do about it is a bit dubious. Obviously, we
don't want to
"fix" .gt() so that it automatically unclasses objects, and I
assume that
zoo has its reasons for not wanting to compare series with different indices. So I suppose that either the user must
unclass, or zoo define
rank.zoo.
Actually qqnorm does not use rank but it does use order
and with the
xtfrm.zoo method I mentioned qqnorm works with zoo;
however, I think
rank needs to be fixed in R to make use of xtfrm as well
since I would
have expected that supplying an xtfrm method for zoo would be
sufficient to
get both order and rank to work without giving errors.
Also note that
rank is not generic.
Notice that xtfrm.default() uses rank().... -- O__ ---- Peter Dalgaard ?ster Farimagsgade 5, Entr.B c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K (*) \(*) -- University of Copenhagen Denmark Ph:
(+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk) FAX:
(+45) 35327907
-- O__ ---- Peter Dalgaard ?ster Farimagsgade 5, Entr.B c/ /'_ --- Dept. of Biostatistics PO Box 2099, 1014 Cph. K (*) \(*) -- University of Copenhagen Denmark Ph:
(+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk) FAX:
(+45) 35327907
***************************************************************** Confidentiality Note: The information contained in this message, and any attachments, may contain confidential and/or privileged material. It is intended solely for the person(s) or entity to which it is addressed. Any review, retransmission, dissemination, or taking of any action in reliance upon this information by persons or entities other than the intended recipient(s) is prohibited. If you received this in error, please contact the sender and delete the material from any computer. *****************************************************************
______________________________________________ R-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel