[R] How to create a function returning an array ?

[Duncan Murdoch]

On 28/06/2013 10:18 AM, Kaptue Tchuente, Armel wrote:
> @ Duncan, I have already used the syntax that you proposed before asking for help by writing something like
> > read_ts<-function(n,m,img) {
> >  out<-.Fortran("read_ts",
> >                as.integer(n),
> >                as.integer(m),
> >                img=as.single(rnorm(n*m)))
> > return(out$img)
> > alpha<-read_ts(n,m)
> > dim(alpha)<-c(n*m)
> > alpha<-t(alpha)
> My worry with this syntax is that (i) the program is not very efficient because n and m are very big and these two additional instructions (dim(alpha)<-c(n*m and alpha<-t(alpha) can be skipped just by directly declaring img as an array in fortran instead of a vector and (ii) the syntax will become more complex dealing with a multidimensional array instead of a matrix as in this example.
> And this is why I'm looking for the correct instruction to declare img as an array instead of a vector.

There are several typos in your code above, but I think your intention 
is clear.

You can do what you are asking for, but not with .Fortran.  It only 
handles vectors for input and output.  You'll need to use .Call (which 
means writing in C or C++).  If you're familiar with C++, using Rcpp is 
probably the easiest way to do this.  If not, I'd rewrite the Fortran 
code to avoid the need for the transpose at the end, and do

dim(out$img) <- c(n,m)
return(out$img)

within your read_ts function.  I think this is reasonably efficient.

Duncan Murdoch


>
> Armel
>
> -----Original Message-----
> From: Duncan Murdoch [mailto:murdoch.duncan at gmail.com]
> Sent: Friday, June 28, 2013 8:16 AM
> To: David Winsemius
> Cc: Kaptue Tchuente, Armel; r-help at r-project.org
> Subject: Re: [R] How to create a function returning an array ?
>
> On 27/06/2013 11:38 PM, David Winsemius wrote:
> > On Jun 27, 2013, at 8:04 PM, Kaptue Tchuente, Armel wrote:
> >
> > > Hi there,
> > >
> > > I would like to know how to change the line
> > > "img=as.single(rnorm(m)))" such that instead of being a vector of
> > > length m as it is now, img is an array of dimension c=(n,m,o) for
> > > instance
> > >
> > > ---------------------------------
> > > read_ts<-function(n,m,o,img) {
> > >   out<-.Fortran("read_ts",
> > >                as.integer(n),
> > >                as.integer(m),
> > >                as.integer(o),
> > >                img=as.single(rnorm(n)))
> > >   return(out$img)
> > > ------------------------------------------
> > >
> >
> > Well, assuming that  the 'out$img' object has a R-length of n*m*o , wouldn't if be simpler to just change the return call to:
>
> In fact, out$img has a length of n, same as on input.  .Fortran won't change the length of its arguments.
>
> Duncan Murdoch
>
> >
> > return( array( out$img, dim=c(n,m,o) )
> >
> > I don't think you wnat start naming your dimension vectors "c".
> >
>