Corrected: approaches to matrix multiplication of each layer of 3-d array with different matrix

What does it mean to unroll the loop?

Thanks!

Ranjan


On Fri, 10 Jul 2015 07:07:15 -0700 Jeff Newmiller
<jdnewmil at dcn.davis.ca.us> wrote:

Strictly speaking, the answer is yes because you can unroll the loop,

but that probably is not what you really want or need to do.

Your example seems about right, but it is not clear how you plan to

make 44 conform with 5500.

---------------------------------------------------------------------------

Jeff Newmiller                        The     .....       .....  Go

Live...

DCN:<jdnewmil at dcn.davis.ca.us>        Basics: ##.#.       ##.#.  Live

Go...

                                      Live:   OO#.. Dead: OO#.. 

Playing

Research Engineer (Solar/Batteries            O.O#.       #.O#.  with
/Software/Embedded Controllers)               .OO#.       .OO#. 

rocks...1k

---------------------------------------------------------------------------

Sent from my phone. Please excuse my brevity.

On July 10, 2015 6:44:19 AM PDT, Ranjan Maitra

<maitra.mbox.ignored at inbox.com> wrote:

Hi,

Sorry to post again, but there is a careless error in my first R

code

snippet:

--- begin R code ---

AA <- array(1:60, dim = c(5500,44,33))
BB <- array(1:60, dim = c(44,44,33))


arraymatprod <- function(A, B) {
   EE <- array(dim=dim(A));
   for (i in 1:dim(A)[3]) EE[,,i] <- A[,,i] %*% B[,,i]
   EE
}

system.time(arraymatprod(AA, BB))

--- end R code ---

The second snippet is correct:

--- begin R code ---

arraymatrixproduct <- function(A, B) {
require(abind)
dA <- dim(A)[1]
AB <- abind(A, B, along = 1)
array(apply(X = AB, MARGIN = 3, FUN = (function(mat) ((mat[1:dA, ]

%*%

mat[(dA+1):(dim(AB)[1]),])))), dim = dim(A))
}

system.time(arraymatrixproduct(AA, BB))

--- end R code ---

However, the second is almost twice as long as the first snippet.

Many thanks,
Ranjan


On Fri, 10 Jul 2015 08:23:49 -0500 Ranjan Maitra
<maitra.mbox.ignored at inbox.com> wrote:

Dear friends,

I have two 3-d arrays of appropriate dimensions. I want to

postmultiply the 2-D matrix layers of the first with the 2-D matrix
layers of the second. Can I do this easily in R avoiding loops?

As an example:

--- begin R code ---

AA <- array(1:60, dim = c(5500,44,33))
BB <- array(1:60, dim = c(44,44,33))


arraymatprod <- function(A, B) {
    EE <- array(dim=dim(A));
    for (i in 1:3) EE[,,i] <- A[,,i] %*% B[,,i]
    EE
}

system.time(arraymatprod(AA, BB))

--- end R code ---

So, is there a way to do this without the loop?

Of course, I could abind the arrays and then use apply with an

appropriate function which would be as follows:

--- begin R code ---

arraymatrixproduct <- function(A, B) {
	require(abind)
	dA <- dim(A)[1]
	AB <- abind(A, B, along = 1)
	array(apply(X = AB, MARGIN = 3, FUN = (function(mat) ((mat[1:dA,

]

%*% mat[(dA+1):(dim(AB)[1]),])))), dim = dim(A))

}

system.time(arraymatrixproduct(AA, BB))

--- end R code ---

However, this turns out to be slower -- perhaps because of the use

of

abind and filling the array inside the function.

I just wanted suggestions to get this operation done more

efficiently.

Many thanks and best wishes,
Ranjan