R devel - Jun 2017 - Simplify and By Convert Factors To Numeric Values

Good day,

It's not described anywhere in the help page, but tapply and by functions
will, by default, convert factors into numeric values. Perhaps this needs to be
documented or the behaviour changed.
> tapply(1:3, 1:3, function(x) factor(LETTERS[x], levels = LETTERS))
1 2 3 
1 2 3

The documentation states "... tapply returns a multi-way array containing
the values ..." but doesn't mention anything about converting factors
into integers. I'd expect the values to be of the same type.

R version 3.4.0 (2017-04-21)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Debian GNU/Linux 8 (jessie)

Matrix products: default
BLAS: /dskh/nobackup/biostat/software/R-3.4.0/lib/libRblas.so
LAPACK: /dskh/nobackup/biostat/software/R-3.4.0/lib/libRlapack.so

--------------------------------------
Dario Strbenac
University of Sydney
Camperdown NSW 2050
Australia

On Fri, 16 Jun 2017, Dario Strbenac wrote:
> Good day,
>
> It's not described anywhere in the help page, but tapply and by 
> functions will, by default, convert factors into numeric values. Perhaps 
> this needs to be documented or the behaviour changed.

It *is* described in the help page.

This returns a list of objects and each object class has "factor"

tapply(rep(1:2,2), rep(1:2,2),
   function(x) factor(LETTERS[x], levels = LETTERS))

and this
>
>> tapply(1:3, 1:3, function(x) factor(LETTERS[x], levels = LETTERS))
> 1 2 3
> 1 2 3
returns a vector object with no class.
>
> The documentation states "... tapply returns a multi-way array 
> containing the values ..." but doesn't mention anything about
converting
> factors into integers. I'd expect the values to be of the same type.
and also states

"If FUN returns a single atomic value for each such cell ... and when 
simplify is TRUE ...  if the return value has a class (e.g., an object of 
class "Date") the class is discarded."

which is what just happened in your example.

Maybe you want:

unlist(tapply(1:3, 1:3, function(x) factor(LETTERS[x],
   levels = LETTERS),simplify=FALSE))

Trying to preserve class worked here in a way you might have 
hoped/expected, but might lead to difficulties in other uses.

HTH,

Chuck