the scalar type, i.e. the type of the coefficients
Unlike SparseMatrix, this class provides a much higher degree of flexibility. In particular, it allows random read/write accesses in log(rho*outer_size) where rho is the probability that a coefficient is nonzero and outer_size is the number of columns if the matrix is column-major and the number of rows otherwise.
Internally, the data are stored as a std::vector of compressed vector. The performances of random writes might decrease as the number of nonzeros per inner-vector increase. In practice, we observed very good performance till about 100 nonzeros/vector, and the performance remains relatively good till 500 nonzeros/vectors.
a reference to the coefficient value at given position row, col This operation involes a log(rho*outer_size) binary search. If the coefficient does not exist yet, then a sorted insertion into a sequential buffer is performed.
endFill()
template<typename _Scalar , int _Options, typename _StorageIndex >
use insert() inserts a nonzero coefficient at given coordinates row, col and returns its reference assuming that: 1 - the coefficient does not exist yet 2 - this the coefficient with greater inner coordinate for the given outer coordinate. In other words, assuming *this is column-major, then there must not exists any nonzero coefficient of coordinates ixcol such that i >= row. Otherwise the matrix is invalid.
use insert() Like fill() but with random inner coordinates. Compared to the generic coeffRef(), the unique limitation is that we assume the coefficient does not exist yet.
finalize()
template<typename _Scalar , int _Options, typename _StorageIndex >
