Hi Martin
thanks for this. I see what you say about R not being able to
magically subset the
"accuracy" slot. Which leaves me puzzled as to why anyone would define
a vector class such as "string" (p315). I can't see why "string" is
defined as
it is, rather than use something like "fungi" from V&R chapter 5.
So, my next question is, why does the Green Book recommend
setClass("quaternion", "vector", prototype=numeric())
?
Why not do what the "onion" package does and make a quaternion
a single column of a four-row matrix? What advantage does the
setClass() method above have over the single-column-of-a-four-row-matrix
method?
best wishes
Robin
On 29 Aug 2006, at 11:11, Martin Maechler wrote:
>>>>>> "Robin" == Robin Hankin <[EMAIL PROTECTED]>
>>>>>> on Tue, 29 Aug 2006 10:42:21 +0100 writes:
>
> Robin> In the Green Book, section 7.5 discusses new vector
> classes and uses
> Robin> quaternions
> Robin> as an example of a vector class that needs more than one
> number per
> Robin> element.
>
> Robin> I would like to define a new class that has a numeric
> vector and a
> Robin> logical
> Robin> vector of the same length that specifies whether the
> measurement was
> Robin> accurate.
>
> Robin> The following code does not behave as desired:
>
>>> setClass("thing",representation("vector",accurate="logical"))
> Robin> [1] "thing"
>>> dput(x <- new("thing",1:10,accurate=rep(T,10)))
> Robin> structure(c(1, 2, 3, 4, 5, 6, 7, 8, 9, 10), accurate = c
> (TRUE,
> Robin> TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE, TRUE),
> class =
> Robin> structure("thing", package = ".GlobalEnv"))
>>> x[1:3]
> Robin> [1] 1 2 3
>>> dput(x[1:3])
> Robin> c(1, 2, 3)
>>>
>
> Robin> because, although the "accurate" slot is filled as
> desired in "x",
> Robin> when extracting the first
> Robin> three elements, it seems to be lost.
>
> and you would really expect that ``R'' magically knows to it
> also must subset the accurate slote ?
>
> Robin> What is the appropriate setClass() call to do what I
> want? Or indeed
> Robin> is making "thing"
> Robin> a vector class as sensible idea here?
>
> I think you need to define at least a subset and subassign
> method for your class as well.
>
> Defining it as "vector" will automatically inherit all the
> method definitions for "vector" --- none of which will ``know
> anything''
> about the accuracy slot.
> Therefore, I tend to think you rather define a class with "all slots"
>
> setClass("Thing", representation(x = "numeric", accurate =
> "logical"))
>
> and then you probably have to define many methods for that
> class, notably for "[" and also "[<-" where the latter should
> happen via setReplaceMethod("Thing",
>
> Also, I'd define a validity method, where you have to decide if
> 'accurate' must have the same length as 'x' -- or what it should
> mean if not.
>
> Martin
--
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
