> I was looking at the IOS API and the column generation part is > simple. The > problem is that it is very inconvenient to branch in one of the > variables. It > would mean to fix a independent set an being present or not in > the final > solution. The branches are very unbalanced and, in the case a > column is fixed > in 0, it is hard to make sure that another identical column will > not be > generated. > > Has someone already used a generalized branch with IOS? What do > you think > about a > void ios_generazed_branch(IOS *ios, int ne, double *v, double > rhs, int next); > > that would add the row ( v <= rhs) to one branch and the row (v >>= rhs + 1) to > the other?
You can branch on a constraint, not only on a varieble. In order to do that you should add the constraint to a subproblem (ios allows doing that on any level) and sets up appropriate bounds for every child subproblems. _______________________________________________ Help-glpk mailing list [email protected] http://lists.gnu.org/mailman/listinfo/help-glpk
