Re: [Rd] setMethod("c") [was: setMethod("Summary")]

From: John Chambers <jmc_at_r-project.org>
Date: Wed 06 Sep 2006 - 17:32:18 GMT

You missed the point of the example, which is why your own implementation didn't work.

It's not the tail recursion that is important, but the recasting of max() (or of c()) to not just a standard generic, but to a recursive computation, so that methods need only be defined for a finite number of arguments.

Because the recursion in c() requires two arguments, not one as with max(), the methods are more naturally transferred to an auxiliary function, cPair in my sketch. Then cWithMethods _replaces_ the ordinary c(), it's not just a method for it.

Also required are a set of methods that corresponds to what you want to do. The methods apply, as I said before, to cPair(), which is a generic with two arguments.

If your picture is that you can bind your class to anything, in either order, then you need methods for ("ANY", "brob") and ("brob", "ANY"), as well as the method ("brob", "brob"), equivalent to the function cPairOfBrobs(), and a default method that just uses c().

Something like:



cWithMethods <- function(x, ...) {

   if(nargs()<3)

      cPair(x,...)
    else
      cPair(x, cWithMethods(...))
}

setGeneric("cPair", function(x,y)standardGeneric("cPair"))

setMethod("cPair", c("brob", "brob"), function(x,y)cPairOfBrobs(x,y))

setMethod("cPair", c("brob", "ANY"), function(x,y)c(x@x, y))

setMethod("cPair", c("ANY", "brob"), function(x,y)c(x, y@x))

setMethod("cPair", c("ANY", "ANY"), function(x,y)c(x,y))

Robin Hankin wrote:
> Dear All
>
> thank you for your continued patience and help.
> The example in the Green Book is
>
>
> setGeneric("max",
> function(x, ..., na.rm=FALSE){
> if(nDotArgs(...)>0){
> max(c(max(x, na.rm=na.rm), max(..., na.rm=na.rm)))
> } else {
> standardGeneric("max")
> }
> }
> )
>
>
> The point of this example is to implement a tail recursion. But it
> isn't applicable
> to c() because it is a primitive function and the generic function
> cannot be changed:
>
>
> setGeneric("c",
> function(x, ...){
> z <- list(...)
> if(length(z)>0){
> return(c(x, c(...)))
> } else {
> return(standardGeneric("c"))
> }
> }
> )
>
>
> gives the following error:
>
>
> Error in setGeneric("c", function(x, ...) { :
> 'c' is a primitive function; methods can be defined, but the
> generic function is implicit, and cannot be changed.
>
>
> OK, plan B (or should that be plan A?) is to define cPair() and call
> that .
> Minimal self-contained code follows
>
>
> setClass("brob",
> representation = representation
> (x="numeric",positive="logical"),
> prototype = list(x=numeric(),positive=logical())
> )
>
>
> "brob" <- function(x,positive){
> if(missing(positive)){
> positive <- rep(TRUE,length(x))
> }
> if(length(positive)==1){
> positive <- rep(positive,length(x))
> }
> new("brob",x=x,positive=positive)
> }
>
> is.brob <- function(x){is(x,"brob")}
>
> as.brob <- function(x){
> if(is.brob(x)){
> return(x)
> } else {
> return(brob(log(abs(x)),x>0))
> }
> }
>
>
> cWithMethods <- function(x, ..., recursive=TRUE) {
> if(nargs()<3){
> return(cPairOfBrobs(x, ...))
> } else {
> return(cPairOfBrobs(x, cWithMethods(...)))
> }
> }
>
> cPairOfBrobs <- function(x,y){
> x <- as.brob(x)
> y <- as.brob(y)
> brob(c(x@x,y@x),c(x@positive,y@positive))
> }
>
> setMethod("c",signature("brob"),cWithMethods)
>
>
>
> But this has the same problem as before; if x is a brob,
> then c(x,1) is fine but c(1,x) isn't:
>
>
>
> x <- new("brob",x=pi,positive=T)
> c(x,1)
> An object of class "brob"
> Slot "x":
> [1] 3.141593 0.000000
>
> Slot "positive":
> [1] TRUE TRUE
>
> > c(1,x)
> [[1]]
> [1] 1
>
>
>
>
> How do I tell setMethod("c", ...) to call the appropriate functions
> if any object passed to c()
> is a brob?
>
>
>
>
>
> On 5 Sep 2006, at 16:47, John Chambers wrote:
>
>
>> (Before someone else can embarrass me with the reference)
>>
>> There is a variant on the c() example discussed in "Programming with
>> Data", page 351, for the function max().
>>
>> John
>>
>> John Chambers wrote:
>>
>>> It's all very well to go on about efficiency, but the purpose of
>>> statistical computing is insight into data, not saving CPU cycles (to
>>> paraphrase Dick Hamming).
>>>
>>> S3 methods do some things fine; other tasks need more
>>> flexibility. One
>>> should ask what's important in a particular application and try to
>>> find
>>> tools that match the needs well.
>>>
>>> Now, the c() function. This has been discussed in various forms (and
>>> languages) for some time. As I remember and as far as I know, the
>>> only
>>> really general way to ensure dispatch on _any_ applicable argument
>>> is to
>>> turn the computation into a pair-wise one and define the methods
>>> (NOT S3
>>> methods) for the two arguments of the pairwise function.
>>>
>>> I won't try to reproduce the details off the top of my head (if I
>>> locate
>>> a reference I'll pass it on), but very roughly the idea is to say
>>> something like
>>>
>>> cWithMethods <- function(x, ...) {
>>> if(nargs()<3)
>>> cPair(x,...)
>>> else
>>> cPair(x, cWithMethods(...))
>>> }
>>>
>>> and then write methods for cPair().
>>>
>>> John
>>>
>>> Robin Hankin wrote:
>>>
>>>
>>>> Hello everybody.
>>>>
>>>> I didn't see Franklin's first message; sorry.
>>>>
>>>> Bearing in mind Professor Ripley's comments
>>>> on the efficiency of S4 vs S3, I'm beginning to think I
>>>> should just stick with S3 methods for my brob objects. After
>>>> all, S3 was perfectly adequate for the onion package.
>>>>
>>>> Notwithstanding that, here's my next problem. I want to define a
>>>> brob method for "c". Using the example in package "arules" as a
>>>> template (I couldn't see one in Matrix), I have
>>>>
>>>>
> [snip]
>
>>>> Now, this works for something like
>>>>
>>>>
>>>>> x <- new("brob",x=pi,positive=T)
>>>>> c(x,x)
>>>>>
>>>> but c(1,x) isn't dispatched to my function. How to
>>>> deal cleanly with this case? Perhaps if any argument
>>>> to c() is a brob object, I would like to coerce them all to brobs.
>>>> Is this possible?
>>>>
>>>>
>>>>
>
> --
> Robin Hankin
> Uncertainty Analyst
> National Oceanography Centre, Southampton
> European Way, Southampton SO14 3ZH, UK
> tel 023-8059-7743
>
> ______________________________________________
> R-devel@r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>

        [[alternative HTML version deleted]]



R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel Received on Thu Sep 07 04:55:44 2006

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.1.8, at Thu 07 Sep 2006 - 11:43:01 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-devel. Please read the posting guide before posting to the list.