Suppose I have an optimization problem with 1000 binary variables. I predict the binaries, with say a neural network. With the 1000 predicted binaries, 999 are "correct" in the sense that they belong to a feasible solution, but the last binary is incorrect, i.e. it makes the overall binary variable set infeasible.
Can someone explain what Gurobi does in this situation? Does warm starting with the binaries help at all? Clearly, fixing the binaries will make the problem infeasible, but if I just supply these 1000 binaries as an initial guess... what happens in the branch and bound tree?
Please sign in to leave a comment.