[R] str and structable error

[Martin Maechler]

>>>>> "Petr" == Petr Pikal <petr.pikal at precheza.cz>
>>>>>     on Mon, 28 Nov 2005 08:38:07 +0100 writes:

    Petr> Hallo I encountered a behaviour which puzzles me (but
    Petr> finally I did get what I wanted).


    Petr> I used structable and strucplot but I wanted to change
    Petr> names of variables in structable object. 

structable is not "part of R" :
You should tell us that you are using an extra package where
structable is from : namely -- I did your homework -- "vcd"

    Petr> I tried to subset it, use names but to no avail. So I
    Petr> tried str and expected to get a structure of an object
    Petr> but:

    >> sss<-structable(Titanic)
    >> str(sss)
    Petr> Error in "[.structable"(x, args[[1]], ) : subscript
    Petr> out of bounds

yes, and similar problems from

    library(vcd)
    example(structable)
    str(hec)

--> see below

    Petr> Finally I learned, that I need to change attributes of
    Petr> structable object.

    Petr> Is this error message OK and I did not read
    Petr> documentation properly? Or is it normal that str gives
    Petr> an error on some objects but I just was not so lucky
    Petr> to meet one?.

No, it is not normal (and is a bug -- in vcd code IMO):

The reason is that  structable ``objects'' (S3)
do not fulfill a fundamental property that all  S (and hence R)
objects should fulfill IMO  {but read on before protesting}

      >>>>> length() and "[" should be compatible
      >>>>> -------------------------------------

namely, for an object 'x', if
   n <- length(x)
and assume n > 0 for the moment, 
then
   x[j]
should return something reasonable for all numeric vectors 'j'
which have values in {1,2,...,n}  (and  also for {-n,...,-1})

This is unfortunately not at all true for structable objects.
I'd say the authors of "structable" made a bit a peculiar
decision when they designed the "[.structable" method since that
invalidates the above basic principle.
This is particularly unfortunate, since "structable" also
inherits from "ftable" {a ``standard R'' S3 class} which does
not have that bad property

If the current "[" (non-S-like IMO) behavior of structable
objects really will be maintained in the future, 
one solution / workaround would be to define a simple 
str.structable  method -- which would also help you for the
moment :

str.structable <- function(object, ...) { 
   cat("structable ")
   class(object) <- class(object)[-1]
   str(object, ...) 
}

> str(structable(Titanic))
structable  ftable [1:8, 1:4] 0 118 0 154 35 387 0 670 5 57 ...
 - attr(*, "dnames")=List of 4
  ..$ Class   : chr [1:4] "1st" "2nd" "3rd" "Crew"
  ..$ Sex     : chr [1:2] "Male" "Female"
  ..$ Age     : chr [1:2] "Child" "Adult"
  ..$ Survived: chr [1:2] "No" "Yes"
 - attr(*, "split_vertical")= logi [1:4] FALSE  TRUE FALSE  TRUE
 - attr(*, "col.vars")=List of 2
  ..$ Sex     : chr [1:2] "Male" "Female"
  ..$ Survived: chr [1:2] "No" "Yes"
 - attr(*, "row.vars")=List of 2
  ..$ Class: chr [1:4] "1st" "2nd" "3rd" "Crew"
  ..$ Age  : chr [1:2] "Child" "Adult"
 - attr(*, "class")= chr "ftable"
> 

Regards,
Martin Maechler, ETH Zurich


    Petr> W2000, R2.2.0, vcd package Built: R 2.2.0; ; 2005-11-
    Petr> 22 14:23:44; windows,

    Petr> Best regards.

    Petr> Petr

    Petr> Petr Pikal petr.pikal at precheza.cz