I have a fixed sparse matrix that I set up before my computations through:
Eigen::SparseMatrix<double, Eigen::ColMajor> A;
std::vector<Eigen::Triplet<double>> coefficients{
{0, 0}//, {,},{,},{,}
};
A.setFromTriplets(coefficients.begin(), coefficients.end());
A.makeCompressed();
Now, I need to update it with my actual entries. Unfortunately, going through the entries with:
for(...) A.coeffRef(i,j) = new_val(i,j);
I get a binary search instead of an efficient constant time access.
I could use valuePtr()
, but it seems a little bit "hacky".
One way is to go with iterators and update the entries with them:
for (int k=0; k<A.outerSize(); ++k)
for (SparseMatrix<double>::InnerIterator it(A,k); it; ++it)
it.valueRef() = new_val(it.row(), it.col());
In my case, I can unroll the loops and populate manually each entry in code:
A.coeffs()[0] = new_val_0;
A.coeffs()[1] = new_val_1;
A.coeffs()[2] = new_val_2;
...
Be aware that this is a column-major sparse matrix, so it's fast j
and slow i
. Populating the coefficients by hand requires to keep this fact in mind.