Development of a Parametric Generating Procedure for Integer Programming Test Problems

Abstract

Integer programming researchers usually test the efficiency of proposed algorithms by solving a few randomly generated problems with different problem sizes and densities However, there is reason to believe that primal and dual degeneracy tn the linear relaxation are also important parameters of problem difficulty, and theoretical results for cutting-plane procedures indicate that the determinant of the linear programming optimal basis ts another relevant parameter This paper presents a procedure that permits researchers to generate feasible random test problems with specified values of these and related problem parameters KEY WORDS AND PHRASES integer programming, experimental design, statistics, problem generation CR CAIEGORIES. 5 25, 5 49

Designing and reporting on computational experiments with heuristic methods

“…Generators typically give the user control over problem characteristics Arthur & Frendewey, 1994;Klingman et al, 1974 , thereby allowing creation of instances with a speci c con guration of experimental design factors Klingman & Mote, 1987 . Some codes supply an optimal solution or its value Arthur & Frendewey, 1988;Lin & Rardin, 1977 . If written for machine independence and portability, generators also provide an e cient means of distributing and reproducing problems.…”