Multiple Observations and Goodness of Fit in Generalized Inverse Optimization
Timothy C. Y. Chan, Taewoo Lee, Rafid Mahmood, Daria Terekhov · arXiv (Cornell University) · 2018
This paper develops a generalized inverse linear optimization framework for imputing objective function parameters given a data set containing both feasible and infeasible points. We devise assumption-free, exact solution methods to solve the inverse problem; under mild assumptions, we show that these methods can be made more efficient. We extend a goodness-of-fit metric previously introduced for the problem with a single observed decision to this new setting, proving and numerically illustrating several important properties.