
Boost : 
From: Benedikt Weber (weber_at_[hidden])
Date: 20020623 13:11:27
I was playing around with ublas today to better understand how it works. One
point that got my attention is the 2dimensional iterator (officially
"iterator1" and "iterator2"). I was assuming that the iterators of a sparse
matrix iterate over the nonzero elements, but when trying it out, I found
the following:
numerics::sparse_matrix<double> A(3,3,1);
A(0,0)= 1;
for (numerics::sparse_matrix<double>::iterator1 i= A.begin1(); i !=
A.end1(); ++i)
for (numerics::sparse_matrix<double>::iterator2 j= i.begin(); j !=
i.end(); ++j)
std::cout << "(" << i.index1() << "," << j.index2() << ") ";
std::cout << std::endl;
This prints out:
(0,0) (1,0) (1,1) (1,2) (2,0) (2,1) (2,2)
Is this intention and what would the reason, or is it a bug?
The consequence is that matrix multiplication takes very long for sparse
matrices. The following code takes forever, even though there is only one
multiplication involved.
// large matrices with just one entry
numerics::sparse_matrix<double> B(1000,1000,1);
numerics::sparse_matrix<double> C(1000,1000,1);
B(0,0)= 1;
C(0,0)= 1;
numerics::sparse_matrix<double> D= numerics::prod(B,C);
I certainly hope this is an easy to fix implementation bug or a wrong
interpretation on my side and not a design problem. (sparse_vector works a
expected and so is the sparse_matrix/sparce_vector multiplication)
Benedikt Weber
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk