A Visual Web Tool to Perform What-If Analysis of Optimization Approaches


In Operation Research, practical evaluation is essential to validate the efficacy of optimization approaches. This paper promotes the usage of performance profiles as a standard practice to visualize and analyze experimental results. It introduces a Web tool to construct and export performance profiles as SVG or HTML files. In addition, the application relies on a methodology to estimate the benefit of hypothetical solver improvements. Therefore, the tool allows one to employ what-if analysis to screen possible research directions, and identify those having the best potential. The approach is showcased on two Operation Research technologies: Constraint Programming and Mixed Integer Linear Programming.
Submitted 16 Mar 2017 to Artificial Intelligence [cs.AI]
Published 20 Mar 2017
Subjects: cs.AI cs.PF