[R] "may be used in an incorrect context" [solved]

[Ross Boylan]

On Tue, 2011-07-19 at 13:55 +1200, Rolf Turner wrote:
> I can't give you a definitive answer, but your use of "..." in your call 
> to optim()
> makes no sense at all to me.  The "..." argument to optim() is used to 
> pass further
> arguments (in addition to the argument over which the objective function is
> being optimized) to the objective function.  If these "further 
> arguments" are to
> be specified in a "..." argument to the function which is actually 
> calling optim()
> then your call to optim() *could* make sense.
Just the hint I needed!  Thank you.  The problem is actually not further
down (the inner function being optimized over has no ...), but further
up.

The arguments I wanted to capture were control parameters for optim
itself.  However, the outer function had no ... argument, and so it was
not a meaningful reference.

I added ... to the outer function, which in turn revealed that, for
various reasons, I wasn't going to be able to pass arguments down that
way (e.g., some of the arguments I wanted to pass were already named
arguments--with different meanings--in some of the functions through
which the ... was passed.

Although I haven't gotten things to run with the added ..., the warnings
(which became runtime errors) went away.

Here's a small example:

foo <- function(a, ...) {
  a+10
}

bar <- function(b){
  b*foo(3, ...)
}

> bar(4)
Error in bar(4) : '...' used in an incorrect context
The body of bar references ..., but it is not in the formal parameter
list of bar.

bar <- function(b, ...)
would make the reference to ... in the body sensible.

Ross

> 
> Except for the fact that your objective function --- agfitfn() --- has 
> no further
> arguments at all!  It is a function of beta only.  (Whence I suspect 
> that there
> is no "..." argument in the calling function and that you are putting in 
> "..."
> in the call to optim() in a mindless manner with no idea of what the 
> inclusion
> of this argument is supposed to accomplish.  Feel free to demonstrate that
> I am wrong in my suspicion.)
> 
> If your objective function (here agfitfn()) actually had further 
> arguments (say "melvin",
> "clyde", and "irving" and if your calling function had usage of the form
> 
>      foo(this,that,theOther,...)
> 
> then you might call foo() in the manner
> 
>      gorp <- 
> foo(17,42,xxx,melvin=someThing,clyde=someThingElse,irving=whatEver)
> 
> where xxx, someThing, someThingElse and whatEver are previously defined 
> objects.  Then
> inside foo() a call to optim() of the form
> 
>      r <- optim(init,agfitfn,...)
> 
> would be sensible.  The value of "..." would be the (named) list with 
> components
> melvin=someThing, clyde=someThingElse and irving=whatEver.  This would 
> result
> in agfitfn() being called with arguments beta, melvin=someThing, 
> clyde=someThingElse
> and irving=whatEver, for various values of beta (as optim() searched 
> over various values
> of beta for the optimum).
> 
> Note that in the call to foo() the "..." arguments must be specified in 
> the name=value
> manner.
> 
> But since agfitfn() is a function of beta only, just lose the "..." 
> argument, and your worries
> will vanish.
> 
>      cheers,
> 
>              Rolf Turner
> 
> On 19/07/11 12:48, Ross Boylan wrote:
> > R CMD check tells me
> > * checking R code for possible problems ... NOTE
> > agexact.fit.rds: ... may be used in an incorrect context: ‘optim(init,
> >    agfitfn, ...)’
> > Warning:<anonymous>: ... may be used in an incorrect context: ‘optim(init, agfitfn, ...)’
> >
> > Can anyone tell me what this message means?  My searches haven't turned
> > up anything useful.  This is with R 2.7 and 2.9.
> >
> > The message appears to be referring to a section of code whose
> > highlights are
> >      agfitfn<- function(beta) {
> >        r<- .C("agexactrds", iter= as.integer(control$iter.max),
> > 		   as.integer(n),
> > 		   as.integer(nvar), sstart, sstop,
> > 		   sstat,
> > 		   x= x[sorted,],
> >                  sextra,
> >                  as.integer(length(response.prob)),
> >                  as.double(response.prob),
> >                  as.double(alpha),
> > 		   as.double(offset[sorted] - mean(offset)),
> > 		   newstrat,
> > 		   means = double(nvar),
> > 		   coef= as.double(beta),
> > 		   u = double(nvar),
> > 		   imat= double(nvar*nvar), loglik=double(2),
> > 		   flag=integer(1),
> > 		   double(2*nvar*nvar +nvar*4 + n),
> > 		   integer(2*n),
> > 		   as.double(control$eps),
> > 		   as.double(control$toler.chol),
> > 		   sctest=double(1),
> >                  as.integer(length(debugfile)),
> >                  as.character(debugfile),
> >                  as.integer(1),  # dumb mode
> >                  PACKAGE="survivalrds" )
> >        - r$loglik[2]
> >      }
> >      r<- optim(init, agfitfn, ...)
> >      return(r)
> >
> > All of this is in the body of
> > agexact.fit.rds<- function([stuff] {
> > 	[more stuff]
> > }
> >
> > Thanks.
> > Ross
> >
> > ______________________________________________
> > R-help at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> > and provide commented, minimal, self-contained, reproducible code.
>