Author:

Abstract:

Problem generators are convenient tools for making large numbers of problem instances available to objectively evaluate the performance of different algorithms. We suggest that a) problem generators should be used only as a last resort, and b) if used they should be 'portable', i.e. , will generate the same problem instances on different computers, and c) use statistical methodology consistent with good experimental design. We provide a number of rules and tools to use when deciding to use a random problem generator.