A Details of Experiments

R is the prize of visited node. Most of MDP is similar with TSP including training scheme. The MDP formulation is mostly same as TSP . This section provides implementation details of the seeder for the experiments. The details of setting T in the inference phase (i.e. in experiments) is described in Appendix A.5. A.3 Detailed Implementation of Reviser This section describes the detailed implementation of the reviser for each target problem.