R devel - Jan 2006 - data.matrix returns mode logical for zero rows (PR#8496)

Full_Name: Jonathan Swinton
Version: 2.2.1
OS: Windows
Submission from: (NULL) (193.132.159.169)


#The first line of description for data.matrix says that it will
#	'Return the matrix obtained by converting all the variables in a
#    data frame to numeric mode and then binding them together as the
#     columns of a matrix.'

#However when called with a data.frame with zero rows, data.matrix returns a
matrix 
#of mode logical rather than numeric. This conflicts with the documentation
#and is not what seems sensible.

# One underlying reason for this is that when a zero-length column of a matrix
of mode logical is 
# asserted to be numeric the matrix is not actually cast to numeric. I wonder if
that too is a bug?
> R.version.string
[1] "R version 2.2.1, 2005-12-20"
> df <- data.frame(matrix(1:2,nrow=2))
> mode(data.matrix(df)[,1])
[1] "numeric"
> mode(data.matrix(df[FALSE,])[,1])
[1] "numeric"
> 
> # Underlying cause
> x <- matrix(nr = 2,nc = 1 )
> mode(x)
[1] "logical"
> x[, 1] <-   c(1,2)
> mode(x)
[1] "numeric"
> 
> x0 <- matrix(nr = 0, nc = 1)
> x0[, 1] <-   numeric(0)
> mode(x0)
[1] "logical"
> mode(x0[,1])
[1] "logical"

On Tue, 17 Jan 2006 jonathan.swinton at astrazeneca.com wrote:
> Full_Name: Jonathan Swinton
> Version: 2.2.1
> OS: Windows
> Submission from: (NULL) (193.132.159.169)
>
>
> #The first line of description for data.matrix says that it will
> #	'Return the matrix obtained by converting all the variables in a
> #    data frame to numeric mode and then binding them together as the
> #     columns of a matrix.'
>
> #However when called with a data.frame with zero rows, data.matrix returns
a
> matrix
> #of mode logical rather than numeric. This conflicts with the documentation
> #and is not what seems sensible.
You don't show us an example of a data frame with zero rows, nor do you 
show an example of data.matrix returning logical, so this report was very 
confusing.  Please do give a reproducible example as the posting guide and 
FAQ ask.

df[FALSE,] is not a data frame, so I don't know why you expect data.matrix 
to be applicable (it is the same as as.matrix in such cases).

Here is an actual reproducible example:

DF <- data.frame(x=c("a", "b"), y=2:3)[FALSE,]
typeof(data.matrix(DF))

> One underlying reason for this is that when a zero-length column of a 
> matrix of mode logical is asserted to be numeric the matrix is not 
> actually cast to numeric. I wonder if that too is a bug?
No, but there is a bug in your description as you are describing 
replacement indexing, not `assertion'.  Coercion in replacement indexing 
is done to accommodate the new values, and there are none in your example.
(That is how S has always done it in my experience, although the Blue 
Book p.359 says otherwise.)

If you want to assert that an object be numeric, use mode(x) <-
"numeric".
>> R.version.string
> [1] "R version 2.2.1, 2005-12-20"
>> df <- data.frame(matrix(1:2,nrow=2))
>> mode(data.matrix(df)[,1])
> [1] "numeric"
>> mode(data.matrix(df[FALSE,])[,1])
> [1] "numeric"
>>
>> # Underlying cause
>> x <- matrix(nr = 2,nc = 1 )
>> mode(x)
> [1] "logical"
>> x[, 1] <-   c(1,2)
>> mode(x)
> [1] "numeric"
>>
>> x0 <- matrix(nr = 0, nc = 1)
>> x0[, 1] <-   numeric(0)
>> mode(x0)
> [1] "logical"
>> mode(x0[,1])
> [1] "logical"
-- 
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