[Rd] (PR#13705) [wishlist, patch] make row() and col() preserve dimnames

[Prof Brian Ripley]

On Sun, 17 May 2009, goodrich at fas.harvard.edu wrote:

> Full_Name: Ben Goodrich
> Version: 2.9.0
> OS: Linux (Debian unstable)
> Submission from: (NULL) (128.103.220.16)
>
>
> row(x), col(x), and functions that call them like lower.tri(x) and 
> upper.tri(x) do not retain the rownames or colnames of x in the 
> matrix that is returned. Example from R version 2.9.0 :
>
> x <- matrix(1:9, nrow = 3, ncol = 3)
> rownames(x) <- LETTERS[1:3]
> colnames(x) <- letters[1:3]
>
> dimnames(row(x)) # NULL
> dimnames(col(x)) # NULL
>
> Is there anyone for whom the expected behavior is to drop the 
> dimnames of x ? It is not consistent with other functions of 
> matrices.

I suspect everyone who reads the help page carefully.

This is not a 'function of a matrix' x but of dim(x) for a 
'matrix-like' object.  The help page makes this clear to me (I am not 
the author), so I think you have misread it.  There is no question of 
'drop the dimames' nor 'retaining the rownames': it is a new object, 
an integer matrix whatever x was.

These functions are mainly for use in programming, and need to be 
efficient -- so adding dimnames that will never be needed by all the 
existing code is a non-trivial overhead.

> By default, row(x) already
> returns the row numbers of x (and similarly for col()), so how would seeing
> integers in the rownames be helpful?
>
> Without patch:
>> row(x)
>     [,1] [,2] [,3]
> [1,]    1    1    1
> [2,]    2    2    2
> [3,]    3    3    3
>
> With patch:
>> row(x)
>  a b c
> A 1 1 1
> B 2 2 2
> C 3 3 3
>
> Patch:
> Index: src/library/base/R/matrix.R
> ===================================================================
> --- src/library/base/R/matrix.R (revision 48553)
> +++ src/library/base/R/matrix.R (working copy)
> @@ -104,8 +104,9 @@
>         labs <- rownames(x, do.NULL=FALSE, prefix="")
>         res <- factor(.Internal(row(dim(x))), labels=labs)
>         dim(res) <- dim(x)
> -        res
> -    } else .Internal(row(dim(x)))
> +    } else res <- .Internal(row(dim(x)))
> +    dimnames(res) <- dimnames(x)
> +    res
> }

Lots of issues here: dimnames() is generic and may not be compatible 
with the object produced, so checks are needed.  And for efficiency 
this should be done in the C-level code that creates the object if 
done at all.

> col <- function(x, as.factor=FALSE)
> @@ -114,8 +115,9 @@
>         labs <- colnames(x, do.NULL=FALSE, prefix="")
>         res <- factor(.Internal(col(dim(x))), labels=labs)
>         dim(res) <- dim(x)
> -        res
> -    } else .Internal(col(dim(x)))
> +    } else res <- .Internal(col(dim(x)))
> +    dimnames(res) <- dimnames(x)
> +    res
> }
>
> crossprod <- function(x, y=NULL) .Internal(crossprod(x,y))
>
> ______________________________________________
> 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