Originally posted by: Sergey Polyakovsky
Dear all,
Assuming that I apply a local search algorithm to every interim solution obtained via CP.next(), could you please advise the best way to introduce the improved solution to the running model. I want to see CP taking advantage of the new upper bound (minimisation problem). I have tried to set the objective equal to a special auxiliary variable via constraint and then minimize the value of the latter, but it immediately degrades the search. In that case, I change the upper bound on the variable as the result of the local search. Using starting points does not work for me well as it seems that the solver simply ignores them. Actually, in this case, I have the solution structure different to the way I encode it within the solver, so I only set a subset of variables.
Can I force the solver to jump to a particular combination of variables and their values? Can I enter my solution to the solutions' store? I generally wonder if CP Optimizer can benefit from such heuristically obtained solutions as CPLEX does via callbacks. In my case, the solver is lost with VNS as does not account for problem's specificity.
Thank you!
#CPOptimizer#DecisionOptimization