• Reformulation in the large:
  • First stage of the project, mutual reformulations of the pure VRP and JSP and application of the standard solving techniques:
    • Default vehicle routing encoding (ILOG Dispatcher 3.0, see ILOG Dispatcher 3.0 User's Manual);
    • Vehicle routing reformulated as open shop (ILOG Scheduler 5.0);
    • Default job shop encoding (ILOG Scheduler 5.0, see ILOG Scheduler 5.0 User's Manual);
    • Job shop reformulated as vehicle routing.
  • The second stage of the project, more efficient code:
    • Default vehicle routing encoding (ILOG Dispatcher 3.1, see ILOG Dispatcher 3.1 User's Manual); differs from that above in particular modelling issues to do with software version differences; includes a subclass extention to the IloDistance standard class; the extention allows one to specify distances between nodes as a matrix.
    • Vehicle routing reformulated as open shop (ILOG Scheduler 5.1); differences from the above encoding include a more efficient way of dealing with capacities; a clearer model; search made simpler;
    • Job shop reformulated as vehicle routing (for Dispatcher 3.1); a better search procedure, iterative approach that consists in setting a soft deadline and gradually improving it.
  • The third stage of the project, a large-scale symmetric experimental study of the influence of the conceptual differences between the VRP and job shop. We start off from both pure VRP and JSP and varying one problem parameter at a time move across the VRP-JSP continuum analysing algorithmic performance:
    • Problem generator that produces pure JSPs, pure VRPs or mixed VRPs and JSPs. This code needs a data file with post codes and x/y-coordinates of geographical locations; the generator also needs lists.cpp and link1.h to maintain the necessary data structures.
    • (mixed) VRPs are encoded as flexible job shop for ILOG Scheduler 5.2.
    • (mixed) VRPs are encoded as VRPs for ILOG Dispatcher 3.2.
• Reformulation in the small:
  • Maximal Independent Sets Problem (a Solver 5.0 encoding);
  • Balanced Incomplete Block Design Problem (a Choco encoding).
• Some examples to illustrate the usage of different concepts of search a la ILOG Solver 5.x and Scheduler 5.x:
  • Solver Example 1: shows how you could specify value orderings, in the example random value ordering is used, you could change it adjusting it to your needs;
  • Solver Example 2: shows how you could specify variable orderings, in the example shortest-domain-first variable ordering is used, you could adjust it to your needs;
  • Solver Example 3: shows how you could specify both value & variable orderings at the same time;
  • Scheduler Example: shows how you could specify different resource allocation/scheduling heuristics: interleaving allocation-sequencing phases using texture measurements, random possible resource assignments (when a resource constraint has a few alternative resources), min-number-of-possible-resources choice of resource constraints. The code uses a data file test.dat, which is a 5x5 job shop instance in a special format (see the readme file in the PRAS VRP/JSP Problem Instances directory of the Resources section of this site).

For code, also see the Resources tab of this page.