A learning-based algorithm to quickly compute good primal solutions for Stochastic Integer Programs

We propose a novel approach using supervised learning to obtain near-optimal\nprimal solutions for two-stage stochastic integer programming (2SIP) problems\nwith constraints in the first and second stages. The goal of the algorithm is\nto predict a "representative scenario" (RS) for the problem such that,\ndeterministically solving the 2SIP with the random realization equal to the RS,\ngives a near-optimal solution to the original 2SIP. Predicting an RS, instead\nof directly predicting a solution ensures first-stage feasibility of the\nsolution. If the problem is known to have complete recourse, second-stage\nfeasibility is also guaranteed. For computational testing, we learn to find an\nRS for a two-stage stochastic facility location problem with integer variables\nand linear constraints in both stages and consistently provide near-optimal\nsolutions. Our computing times are very competitive with those of\ngeneral-purpose integer programming solvers to achieve a similar solution\nquality.\n