R devel - Nov 2006 - Matrix*vector: coercing sparse to dense matrix for arithmetic

Hi,

I have a sparse Matrix (kronecker product of spline design matrices),
and I need to multiply each row by a number to get another matrix.  If
the matrix is A and the numbers are stored in a vector k, with plain
vanilla matrices I would do

A*k

But when using the Matrix package (class of A is "dgCMatrix"), I get
the warning "coercing sparse to dense matrix for arithmetic".  The
error message is perfectly reasonable, I am looking for a way to do it
right (keeping operations in the realm of sparse matrices).  Any help
would be appreciated.

Thanks,

Tamas

On 11/29/06, Tamas K Papp <tpapp at princeton.edu> wrote:
> I have a sparse Matrix (kronecker product of spline design matrices),
> and I need to multiply each row by a number to get another matrix.  If
> the matrix is A and the numbers are stored in a vector k, with plain
> vanilla matrices I would do
> A*k
> But when using the Matrix package (class of A is "dgCMatrix"), I
get
> the warning "coercing sparse to dense matrix for arithmetic". 
The
> error message is perfectly reasonable, I am looking for a way to do it
> right (keeping operations in the realm of sparse matrices).  Any help
> would be appreciated.
I would suggest premultiplying by a diagonal matrix.  However, when I
look at the method dispatch after an example it doesn't appear to be
making a good choice.  What I think should work well (and Martin or I
will make the necessary changes to ensure that it does) is

library(Matrix)
data(KNex); mm <- KNex$mm
system.time(Diagonal(x = rnorm(nrow(mm))) %*% mm)