There are lots of references about Boost::uBLAS compressed_matrix and coordinate_matrix. But I found no explanation about mapped_matrix. How is it implemented?
Which strategy of sparse storage does it use?
Plus: can someone provide me some reference in papers or books?
There doesn't seem to be actual books for this library. The documentation is here
http://www.boost.org/doc/libs/1_66_0/libs/numeric/ublas/doc/matrix_sparse.html#mapped_matrix
On quick scan it answers your questions:
It's implemented as a container adapter. It only stores non-zero elements. These are stored in an associative container internally.
By default the underlying associative container is
std::map, but it's customizable using theAtemplate argument.It only stores non-zero elements. These are stored in an associative container internally. See above.
Additionally, you can note the precise way in which the element indices are transformed into a key for the associative container (depending on
F: row-major or column-major organization) and what guarantees this holds for weak total ordering of elements in a sparse matrix compared to the corresponding dense matrix models. (See above).