Neighborhood Synthesis from an Ensemble of MIP and CP Models
In this paper we describe a procedure that automatically synthesizes a neighborhood from an ensemble of Mixed Integer Programming (MIP) and/or Constraint Programming (CP) models. We move on from a recent paper by Adamo et al. (2015) in which a neighborhood structure is automatically designed from a (single) MIP model through a three-step approach: (1) a semantic feature extraction from the MIP model; (2) the derivation of neighborhood design mechanisms based on these features; (3) an automatic configuration phase to find the “proper mix” of such mechanisms taking into account the instance distribution. Here, we extend the previous work in order to generate a suitable neighborhood from an ensemble of MIP and/or CP models of a given combinatorial optimization problem. Computational results show relevant improvements over the approach considering a single model.
KeywordsCombinatorial optimization Neighborhood search Automatic neighborhood design Feature extraction
- Adamo, T., Ghiani, G., Grieco, A., Guerriero, E., Manni, E.: MIP neighborhood synthesis through semantic feature extraction and automatic algorithm configuration. Comput. Oper. Res. (2015, submitted). http://www.emanuelemanni.unisalento.it/Sito/Papers/MIPneighSynth.pdf
- Cordeau, J.-F., Laporte, G., Savelsbergh, M.W., Vigo, D.: Vehicle routing. In: Barnhart C., Laporte, G. (eds.) Transportation. Handbook in OR & MS, vol. 14, pp. 367–428 (2006)Google Scholar
- Hutter, F., Hoos, H., H., Stützle, T.: Automatic algorithm configuration based on local search. In: Proceedings of AAAI-07, vol. 22, pp. 1152–1157 (2007)Google Scholar