We study a class of non-convex sum-of-ratios programs which can be used for decision-making in prominent areas such as product assortment and price optimization, facility location, and security games. Such an optimization problem involves both continuous and binary decision variables and is known to be highly non-convex and intractable to solve. We explore a discretization approach to approximate the optimization problem and show that the approximate program can be reformulated as mixed-integer linear or second-order cone programs, which can be conveniently handled by an off-the-shelf solver (e.g., CPLEX or GUROBI). We further establish (mild) conditions under which solutions to the approximate problem converge to optimal solutions as the number of discretization points increases. We also provide approximation abounds for solutions obtained from the approximated problem. We show how our approach applies to product assortment and price optimization, maximum covering facility location, and Bayesian Stackelberg security games and provide experimental results to evaluate the efficiency of our approach.
