Non-Model-Based Search Guidance for Set Partitioning Problems

We present a dynamic branching scheme for set partitioning problems. The idea is to trace features of the underlying MIP model and base search decisions on current subproblem be solved. show how such system can trained efficiently by introducing minimal learning bias that traditional model-based machine approaches rely on. Experiments highly heterogeneous collection instances significant gains over guidance in Cplex as well instance-specifically tuned pure heuristics.