[Rd] Problem with _new_ if class "lm" in object representation.
John Chambers
jmc at research.bell-labs.com
Wed Sep 29 21:34:02 CEST 2004
Witold Eryk Wolski wrote:
>
> Hi,
> The hint to use an lm0 instance in the protopy is a solution.
>
> But I thought about a different solution to which I got used. This is to
> declare class "lm" as an S4 class by setClass("lm").
It's a good idea to declare S3 classes, but that's not the way to do
it. If you want to use an S3 class as an S4 class, that's what
setOldClass() is for. See the green book or the documentation.
Why doesn't that work for "lm"? As the error message says below, "lm"
is already defined in this way & sealed, just as other standard classes
are sealed.
This is NOT a solution to your previous problem, however, since it's
still totally unknown what a default object from an S3 class should be
like.
You will still get an error from new("Ctest", test = "bla") if you have
an undefined slot from an S3 class, or any other VIRTUAL class. (It
happens that new("Ctest") doesn't fail because if there are no arguments
to new() other than the class name, the validity method isn't called.)
Your "working" solution doesn't work, as in the following simpler
example:
R> setClass("bar")
[1] "bar"
R> setClass("foo", representation(x="numeric", y="bar"))
R> new("foo", x=1)
Error in validObject(.Object) : Invalid "foo" object: Invalid object for
slot "y" in class "foo": got class "NULL", should be or extend class
"bar"
(We should in fact warn on the call to setClass() in this case, since
the protoype object is not a valid object from the class. For 2.0.1)
> This is because I
> have to do it anyway quite frequently when using S3 classes, declared
> in contributed packages, in S4 classes.
> Please consider the following code with this quite recent version
> generating a warning.
> R : Copyright 2004, The R Foundation for Statistical Computing
> Version 2.0.0 beta (2004-09-20), ISBN 3-900051-07-0
>
> #I like to have an object which contains an object of class fields:Krig.
>
> library(fields)
>
> setClass("Ctest"
> ,representation(
> test="character"
> ,bla="character"
> ,mod="Krig"
> )
> )
>
> #Warning message:
> #Undefined slot classes in definition of "Ctest": mod (class " Krig ") in: .completeClassSlots(ClassDef, where)
>
> The overcome this warning i just modify the code adding
>
> setClass("Krig")
> setClass("Ctest"
> ,representation(
> ....
>
> previous to defining class Ctest.
> This is my working solution. But this one does not work with class "lm".
> Since,
>
> > setClass("lm")
> Error in setClass("lm") : "lm" has a sealed class definition and cannot
> be redefined
>
> I understand the errror message but I can't see the difference between
> class Krig and lm which are both S3 classes. (of course except that lm
> is in in THE stats package)
> Is it intended to keep two different "models" of treating S3 classes?
> One for S3 class from the recomended, "base" packages and a different
> one for S3 classes from contributed packages? I assume not.
> If, so I feel a alienated because my code so far rely on the behaviour
> which I observed for S3/contributed classes.
>
> /E,
>
> John Chambers wrote:
>
> >Wolski wrote:
> >
> >
> >>Hi!
> >>Consider this code.
> >>
> >>setClass("Ctest"
> >> ,representation(
> >> test="character"
> >> ,bla="character"
> >> ,mod="lm"
> >> )
> >> )
> >>
> >>new("Ctest",test="bla") #This produces an error.
> >>
> >>#Error in validObject(.Object) : Invalid "Ctest" object: Invalid object for slot "mod" in class "Ctest": got class "NULL", should be or extend class "lm"
> >>
> >>
> >
> >And the error message is correct. Since "lm" is an S3 class, there is
> >no way to generate an object automatically from it. new("lm") will also
> >produce an error.
> >
> >If you want objects to be generated from your class with a default "lm"
> >object, you have to first decide what that is. Assuming you know, you
> >need to either put that element into the protoype for the class, or have
> >a method for initialize() that does something sensible, using the rest
> >of the data in the object.
> >
> >The prototype solution could look something like:
> >
> >R> lm0 = lm(1~1)
> >R> setClass("Ctest", representation(test = "character",
> > bla = "character", mod = "lm"), prototype=prototype(mod = lm0))
> >R> new("Ctest", test="bla")
> >An object of class "Ctest"
> >Slot "test":
> >[1] "bla"
> >
> >Slot "bla":
> >character(0)
> >
> >Slot "mod":
> >
> >Call:
> >lm(formula = 1 ~ 1)
> >
> >Coefficients:
> >(Intercept)
> > 1
> >
> >
> >There is no automatic relation between the class "lm" and the function
> >lm(). The green book recommends having a generator function, but the R
> >code can't assume this has been done.
> >
> >
> >
> >>setClass("Ctest"
> >> ,representation(
> >> test="character"
> >> ,bla="character"
> >> ,mod="character" # its the only with the class definition above.
> >> )
> >> )
> >>new("Ctest",test="bla") #this works as I would expect.
> >>
> >>#An object of class "Ctest"
> >>#Slot "test":
> >>#[1] "bla"
> >>#
> >>#Slot "bla":
> >>#character(0)
> >>#
> >>#Slot "mod":
> >>#character(0)
> >>
> >>Thought that this is due to that the class lm has no lm(0) object. Hence i tried
> >>
> >>setClass("brum",representation(brum="brum"))
> >>setClass("Ctest"
> >> ,representation(
> >> test="character"
> >> ,bla="character"
> >> ,mod="brum" # its the only with the class definition above.
> >> )
> >> )
> >>
> >>new("Ctest",test="best") #but this works to.
> >>
> >>______________________________________________
> >>R-devel at stat.math.ethz.ch mailing list
> >>https://stat.ethz.ch/mailman/listinfo/r-devel
> >>
> >>
> >
> >
> >
>
> --
> Dipl. bio-chem. Witold Eryk Wolski
> MPI-Moleculare Genetic
> Ihnestrasse 63-73 14195 Berlin _
> tel: 0049-30-83875219 'v'
> http://www.molgen.mpg.de/~wolski / \
> mail: witek96 at users.sourceforge.net ---W-W----
> wolski at molgen.mpg.de
--
John M. Chambers jmc at bell-labs.com
Bell Labs, Lucent Technologies office: (908)582-2681
700 Mountain Avenue, Room 2C-282 fax: (908)582-3340
Murray Hill, NJ 07974 web: http://www.cs.bell-labs.com/~jmc
More information about the R-devel
mailing list