R devel - Jan 2004 - Naming difference in cbind between S-Plus (PR#6515)

Naming difference in cbind between S-Plus and R.
I think R is wrong.

abc <- data.frame(y=1:4, x=rnorm(4))
abc.lm <- lm(y ~ x, data=abc)
predict.lm(abc.lm, type="terms")  ## this is where R got the name
"x"
abc <- cbind(abc, d=abc$y - predict.lm(abc.lm, type="terms"))
abc


R gives
> abc
  y           x        x
1 1 -1.33925477 1.163001
2 2  1.52764505 1.799073
3 3  0.07805628 2.983086
4 4 -0.48720310 4.054840

S-Plus gives
> abc
  y          x        d 
1 1  1.8843692 2.075596
2 2 -0.9144517 1.296060
3 3  0.5063175 3.199408
4 4 -0.7054668 3.428936


Rich

--please do not edit the information below--

Version:
 platform = i386-pc-mingw32
 arch = i386
 os = mingw32
 system = i386, mingw32
 status = 
 major = 1
 minor = 7.1
 year = 2003
 month = 06
 day = 16
 language = R

Windows XP Home Edition (build 2600) Service Pack 1.0

Search Path:
 .GlobalEnv, package:methods, package:ctest, package:mva, package:modreg,
package:nls, package:ts,
file:c:/HOME/rmh/hh/splus.library/.RData, package:grid, package:lattice,
Autoloads, package:base

On Sat, 31 Jan 2004 22:27:21 +0100 (CET), you wrote:
>Naming difference in cbind between S-Plus and R.
>I think R is wrong.
I'm not sure if R is right or wrong, but I suspect the difference
isn't in cbind, it's elsewhere...
>abc <- data.frame(y=1:4, x=rnorm(4))
>abc.lm <- lm(y ~ x, data=abc)
>predict.lm(abc.lm, type="terms")  ## this is where R got the name
"x"
>abc <- cbind(abc, d=abc$y - predict.lm(abc.lm, type="terms"))
... in the line above.  In S-PLUS, what does

abc$y - predict.lm(abc.lm, type="terms")

give?  In R it gives a matrix with the column name "x".  It makes
sense to me that using cbind on a matrix with named columns should
keep those column names.  

What does S-PLUS do in this case:

M <- matrix(1:4, 2, 2, dimnames=list(NULL, c('a', 'b')))
cbind(c=5:6,d=M)

R gives 
> M <- matrix(1:4, 2, 2, dimnames=list(NULL, c('a', 'b')))
> cbind(c=5:6,d=M)
     c a b
[1,] 5 1 3
[2,] 6 2 4

which seems reasonable to me, though I might like the name "d" to be
incorporated somehow.

Duncan Murdoch

>From ?cbind
     For 'cbind' ('rbind') the column (row) names are taken from
the
     names of the arguments, or where those are not supplied by
     deparsing the expressions given (if that gives a sensible name).
     The names will depend on whether data frames are included: see the
     examples.

and you cbind-ed a data frame with columns y, x to a data frame with
column x.  It seems to me that what S-PLUS does is equally inconsistent:

abc <- data.frame(y=1:4, x=rnorm(4))
d <- data.frame(x=1:4, e=5:8)
> cbind(abc, d)
  y          x x e
...
> cbind(abc, d=d)
  y          x d.x d.e
...
> cbind(abc, d=d[1])
  y          x d
...
> cbind(abc, d[1])
  y          x x
...

and I would have expected the third to give the name d.x, not d and not x.

Differences from S-PLUS (and please learn to write that correctly!) are 
not per se bugs in R.


On Sat, 31 Jan 2004 rmh@temple.edu wrote:
> Naming difference in cbind between S-Plus and R.
> I think R is wrong.
> 
> abc <- data.frame(y=1:4, x=rnorm(4))
> abc.lm <- lm(y ~ x, data=abc)
> predict.lm(abc.lm, type="terms")  ## this is where R got the name
"x"
> abc <- cbind(abc, d=abc$y - predict.lm(abc.lm, type="terms"))
> abc
> 
> 
> R gives
> > abc
>   y           x        x
> 1 1 -1.33925477 1.163001
> 2 2  1.52764505 1.799073
> 3 3  0.07805628 2.983086
> 4 4 -0.48720310 4.054840
> 
> S-Plus gives
> > abc
>   y          x        d 
> 1 1  1.8843692 2.075596
> 2 2 -0.9144517 1.296060
> 3 3  0.5063175 3.199408
> 4 4 -0.7054668 3.428936
> 
> 
> Rich
> 
> --please do not edit the information below--
> 
> Version:
>  platform = i386-pc-mingw32
>  arch = i386
>  os = mingw32
>  system = i386, mingw32
>  status = 
>  major = 1
>  minor = 7.1
>  year = 2003
>  month = 06
>  day = 16
>  language = R
> 
> Windows XP Home Edition (build 2600) Service Pack 1.0
> 
> Search Path:
>  .GlobalEnv, package:methods, package:ctest, package:mva, package:modreg,
package:nls, package:ts,
> file:c:/HOME/rmh/hh/splus.library/.RData, package:grid, package:lattice,
Autoloads, package:base
> 
> ______________________________________________
> R-devel@stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-devel
> 
> 
-- 
Brian D. Ripley,                  ripley@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

I see valid reasons and will therefore adopt the defensive coding practice:
  abc <- cbind(abc, d=abc$y - predict.lm(abc.lm, type="terms")[,1])
which works the same in both S-Plus and R and gives the answer I want.

The result of predict.lm is not a data.frame.
It is a "structure" in S-Plus and a "matrix" in R.

S-Plus gives
> M <- matrix(1:4, 2, 2, dimnames=list(NULL, c('a', 'b')))
> cbind(c=5:6,d=M)
     c a b 
[1,] 5 1 3
[2,] 6 2 4



On spelling of the name S-Plus or S-PLUS, both are correct.
I normally write in LaTeX, so I normally use {\sc S-Plus}.

First, from the literature,
the Proprietary Notice on the copyright page of the July 2001 printed
edition says
      {\sc S-Plus} 6 for Windows User's Guide

Most of the internal uses of the word in the book are also in
smallcaps {\sc S-Plus}.  The outside cover does say S-PLUS.  The title
page says {\sc S-Plus}.

Second, by personal discussion,
I discussed this spelling with Tim Hesterberg about a year ago.
He told me that S-Plus is an appropriate spelling.

Third, by gestalt,
"S-PLUS" looks like the "PLUS" is the most important
part---simply because
it occupies more space.
"S-Plus" looks like the "S" is the most important
part---because the "S" is
full size and the "Plus" is typographically subsidiary.


Rich