Q: replacing elements in an array

Dear all,


I have a question about the way the replacement of elements in an array
works in R (having a slight idea of the answer), and a question about
the way to achieve my goals.

If I have a 3 ways-array called 'a', the command

a[1, , ] <- rnorm(10)  # assumes the size matches

will result in having the whole object 'a' copied (since arrays are
just 'vectors with dimensions' in R) ... and at some point
two objects like 'a' will be in memory, right ? (this was the first
question).

A <- array(0, dim=c(500, 500, 10))
object.size(A)

memory.size(T)

A[1,,] <- 1
memory.size(T)

I have rather large arrays, and I would like to modify some of their
elements in-situ (to save memory). Is there a way to do that ?

Many thanks for the details... however I could find much about
the function 'memory.size' you use in your example (see below)...
is it a new function only in R-devel ?

On Tue, 13 Aug 2002, Laurent Gautier wrote:

Dear all,


I have a question about the way the replacement of elements in an array
works in R (having a slight idea of the answer), and a question about
the way to achieve my goals.

If I have a 3 ways-array called 'a', the command

a[1, , ] <- rnorm(10)  # assumes the size matches

will result in having the whole object 'a' copied (since arrays are
just 'vectors with dimensions' in R) ... and at some point
two objects like 'a' will be in memory, right ? (this was the first
question).

I don't think so.  See do_subassign and descendants.  It contains

    /* If there are multiple references to an object we must */
    /* duplicate it so that only the local version is mutated. */
    /* This will duplicate more often than necessary, but saves */
    /* over always duplicating. */
    /* FIXME: shouldn't x be protected? */

    if (NAMED(CAR(args)) == 2)
	x = SETCAR(args, duplicate(CAR(args)));

Now coercion can force a copy, but starting from a numeric array I think
your example does not duplicate.  And I can prove it:

A <- array(0, dim=c(500, 500, 10))
object.size(A)

[1] 20000128

memory.size(T)

[1] 29028352

A[1,,] <- 1
memory.size(T)

[1] 29028352

so there was never a second copy.

I have rather large arrays, and I would like to modify some of their
elements in-situ (to save memory). Is there a way to do that ?

See above.  Do your own experiments / reading of code?

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595