R devel - Dec 2007 - bug in by.data.frame, R-2.6.1 (PR#10506)

by() fails for 1-column matrices and dataframes:

X <- data.frame(a=1:10)
g <- gl(2,5)
by(X, g, colMeans)


Suggested fix:

--- by-old.R    2007-12-10 15:26:22.501086600 +0100
+++ by.R        2007-12-10 15:25:58.390477200 +0100
@@ -26,7 +26,7 @@
          IND[[1]] <- INDICES
          names(IND) <- deparse(substitute(INDICES))[1]
      } else IND <- INDICES
-    FUNx <- function(x) FUN(data[x,], ...)
+    FUNx <- function(x) FUN(data[x, , drop=FALSE], ...)
      nd <- nrow(data)
      ans <- eval(substitute(tapply(1:nd, IND, FUNx)), data)
      attr(ans, "call") <- match.call()



                _
platform       i386-pc-mingw32
arch           i386
os             mingw32
system         i386, mingw32
status
major          2
minor          6.1
year           2007
month          11
day            26
svn rev        43537
language       R
version.string R version 2.6.1 (2007-11-26)


Uwe Ligges

It transpires that package survey relies on the current behaviour (which 
is not new in 2.6.1: R 2.0.0 did it).

The suggested fix is reasonable if 'data' was originally a data frame,
but
the default method promotes vectors to data frames, and the data frame 
method used to drop them back to vectors.

I've backed the fix out of R-patched.

I am not sure if the current undocumented behaviour on vectors is 
100% desirable, but will amend by.default in R-devel to preserve it.

Thomas: I think svymean() needs not to assume that 1-column data frames 
will be dropped.


On Mon, 10 Dec 2007, ligges at statistik.uni-dortmund.de wrote:
> by() fails for 1-column matrices and dataframes:
>
> X <- data.frame(a=1:10)
> g <- gl(2,5)
> by(X, g, colMeans)
>
>
> Suggested fix:
>
> --- by-old.R    2007-12-10 15:26:22.501086600 +0100
> +++ by.R        2007-12-10 15:25:58.390477200 +0100
> @@ -26,7 +26,7 @@
>          IND[[1]] <- INDICES
>          names(IND) <- deparse(substitute(INDICES))[1]
>      } else IND <- INDICES
> -    FUNx <- function(x) FUN(data[x,], ...)
> +    FUNx <- function(x) FUN(data[x, , drop=FALSE], ...)
>      nd <- nrow(data)
>      ans <- eval(substitute(tapply(1:nd, IND, FUNx)), data)
>      attr(ans, "call") <- match.call()
>
>
>
>                _
> platform       i386-pc-mingw32
> arch           i386
> os             mingw32
> system         i386, mingw32
> status
> major          2
> minor          6.1
> year           2007
> month          11
> day            26
> svn rev        43537
> language       R
> version.string R version 2.6.1 (2007-11-26)
>
>
> Uwe Ligges
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
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

On Wed, 19 Dec 2007, Prof Brian Ripley wrote:
> It transpires that package survey relies on the current behaviour (which is
> not new in 2.6.1: R 2.0.0 did it).
>
> The suggested fix is reasonable if 'data' was originally a data
frame, but
> the default method promotes vectors to data frames, and the data frame
method
> used to drop them back to vectors.
>
> I've backed the fix out of R-patched.
>
> I am not sure if the current undocumented behaviour on vectors is 100% 
> desirable, but will amend by.default in R-devel to preserve it.
>
> Thomas: I think svymean() needs not to assume that 1-column data frames
will
> be dropped.
Ok. I think that actually simplifies some things.

  I hope to have some time next week to catch up on a number of package 
issues.

 	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle