[CppAD] CppAD Sparsity patterns
Brad Bell
bradbell at seanet.com
Sat Feb 11 06:08:12 EST 2017
I have long suspected that very large vectors of standard set have a
large overhead and are not an efficient way to represent sparsity
patterns for large matrices.
There is now an alternative sparsity pattern in CppAD
https://www.coin-or.org/CppAD/Doc/sparse_rc.htm
that performs better during the CppAD speed tests. See the heading 02-01
through 02-09 on
https://www.coin-or.org/CppAD/Doc/whats_new_17.htm
I am planning to deprecate the vector of standard sets and boolean
sparsity patterns.
One still has the choice to use vectors of booleans or vectors of sets
during the sparsity calculations; e.g., see internal_bool in
https://www.coin-or.org/CppAD/Doc/for_jac_sparsity.xml#internal_bool
There is an internal data structure , different from vectors of standard
sets, that is specially designed for the sparsity calculations and is
used when internal_bool is false.
More information about the CppAD
mailing list