[Rd] repeat bug report: as.factor argument to row / col fails if NULL dimnames

[Prof Brian Ripley]

On Fri, 12 Dec 2008, Peter Dalgaard wrote:

> Prof Brian Ripley wrote:
>> I have never seen the report in the repository.  There were problems with 
>> R-bugs at around that time (most email submission was stalled for a month 
>> or so).
>
> Maybe, but I don't see a PR# on _any_ of Martin Morgan's posts...
>
>> However, I think the problems are far more serious: the help page says
>>
>>      Returns a matrix of integers indicating their column number in a
>>      matrix-like object.
>> ...
>> 
>> as.factor: a logical value indicating whether the value should be
>>           returned as a factor rather than as numeric.
>> 
>> Value:
>>
>>      An integer matrix with the same dimensions as 'x' and whose
>>      'ij'-th element is equal to 'j'.
>> 
>> There is no mention of using the column labels as the values and the result 
>> was not even a matrix.  I certainly expected the factor levels to be 
>> numeric.
>> 
>> I presume that the intention is to use the column labels (default 
>> seq_len(nrow(x)) ) in place of the numbers, and to return a factor matrix, 
>> so I have implemented that for col() and row().
>> 
>> However, I do wonder that anyone was using as.factor=TRUE given that it did 
>> not do anything like what it was documented to.
>
> There could, I suppose, be applications of the style
>
> library(ISwR)
> dd <- data.frame(count=c(caesar.shoe),
>  caesar=row(caesar.shoe,as.factor=TRUE),
>  size=col(caesar.shoe,as.factor=TRUE)) 
> anova(glm(count~caesar*size,family=poisson,data=dd))
>
> (as.data.frame.table is better in this case, I know). Notice that this will 
> break if you change the value to a factor matrix.
>
> It is not clear that a matrix is desired -- an alternative point of view is 
> that when you request a factor, it is intended to be used along with 
> long-format data.

Yes, it is not clear (and I had given it some thought), but

a) a matrix result was documented and

b) it is easier to remove dims than add them (impossible without the 
original matrix to hand).

>> On Mon, 8 Dec 2008, Dan Davison wrote:
>> 
>>> This bug was reported to r-devel by Martin Morgan on 13th May 2008
>>> (original email pasted below). However I couldn't locate it in the bug
>>> tracker (with searches for row, col, as\.factor) It is still present
>>> in today's svn version (rev 47115). I hope I'm doing something useful
>>> by bringing it up again. I have not sent this to r-bugs, as I'm
>>> concerned that not locating it in the bug tracker could be my failing.
>>>

-- 
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