R help - Oct 2006 - shifting a huge matrix left or right efficiently ?

I'm wondering what's the best way to shift a huge matrix left or right.
My current implementation is the following:

shiftMatrixL <- function(X, shift, padding=0) {
  cbind(X[, -1:-shift], matrix(padding, dim(X)[1], shift))
}

X <- shiftMatrixL(X, 1)*3 + shiftMatrixL(X,2)*5...

However, it's still slow due to heavy use of this function.
The resulting matrix will only be read once and then discarded,
so I believe the best implementation of this function is in C,
manipulating the internal data structure of this matrix.
Anyone know similar package for doing this job ?

Huang-Wen

If you're able to work with the transpose of your matrix, you might 
consider the function 'filter()', e.g.:

 > filter(diag(1:5), c(2,3), sides=1)
Time Series:
Start = 1
End = 5
Frequency = 1
   [,1] [,2] [,3] [,4] [,5]
1   NA   NA   NA   NA   NA
2    3    4    0    0    0
3    0    6    6    0    0
4    0    0    9    8    0
5    0    0    0   12   10
 >

I don't know if the conversion to and from a time-series class will 
impact the timing, but if this might serve your purposes, it's easy to 
do some experiments to find out.

- Tony Plate

Huang-Wen Chen wrote:
> I'm wondering what's the best way to shift a huge matrix left or
right.
> My current implementation is the following:
> 
> shiftMatrixL <- function(X, shift, padding=0) {
>   cbind(X[, -1:-shift], matrix(padding, dim(X)[1], shift))
> }
> 
> X <- shiftMatrixL(X, 1)*3 + shiftMatrixL(X,2)*5...
> 
> However, it's still slow due to heavy use of this function.
> The resulting matrix will only be read once and then discarded,
> so I believe the best implementation of this function is in C,
> manipulating the internal data structure of this matrix.
> Anyone know similar package for doing this job ?
> 
> Huang-Wen
> 
> ______________________________________________
> R-help at stat.math.ethz.ch 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.
>

A little algebra up front can often help.

Note that

 	X <- shiftMatrixL(X, 1)*3 + shiftMatrixL(X,2)*5

(and similarly with more terms on the r.h.s)

is just

 	X <- X %*% mat

where mat is is a matrix of zeroes except that

 	mat[ i+1, i ] == 3
 	mat[ i+2, i ] == 5

and dim(mat) == dim(X).

So forget about explicitly shifting the matrix if you do this in native R 
- just construct a suitable version of mat and use '%*%'. If you must do

this in C shift coefficient vector implicitly using a pointer before 
finding the inner product with each row, and if the matrix is truly large 
follow Tony Plate's advice to transform X first (and look at Chapter 1 of 
Matrix Computations by Golub and Van Loan, 1996, if you need to know why).


On Mon, 9 Oct 2006, Huang-Wen Chen wrote:
> I'm wondering what's the best way to shift a huge matrix left or
right.
> My current implementation is the following:
>
> shiftMatrixL <- function(X, shift, padding=0) {
>  cbind(X[, -1:-shift], matrix(padding, dim(X)[1], shift))
> }
>
> X <- shiftMatrixL(X, 1)*3 + shiftMatrixL(X,2)*5...
>
> However, it's still slow due to heavy use of this function.
> The resulting matrix will only be read once and then discarded,
> so I believe the best implementation of this function is in C,
> manipulating the internal data structure of this matrix.
> Anyone know similar package for doing this job ?
>
> Huang-Wen
>
> ______________________________________________
> R-help at stat.math.ethz.ch 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.
>
Charles C. Berry                        (858) 534-2098
                                          Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	         UC San Diego
http://biostat.ucsd.edu/~cberry/         La Jolla, San Diego 92093-0717