initial condition for MILP solver

[G]

Hi,

how (if possibile) to define initial guess for integer solvers?

[Johan Löfberg]

If you edit definesolvers.m and add

solver(i).supportsinitial = 1;

on line 115 or so, gurobi will accept initial guesses (using assign and the 'usex0' option)

Note though, most often the problem is not finding the optimal solution, but closing the gap from below, so even if you have the optimal solution, it might take gurobi just as long to terminate)

Also note that if you have complex models with high-level operators, YALMIP will not be able to assign initial values for all internal variables

[G]

Ok, but if I have 100 variables, and I use the assign function on 20 of them, this provided guess will be taken into account or not?

[Johan Löfberg]

YALMIP sets all unassigned to 0, so most likely the initial guess is infeasible and thus not useful.

[G]

OK, and i guess that would be the only to accomplish the advice "if you know it, give a good initial solution to the solver"