Tuesday, February 18, 2014

Terrible gaps in MIP model

I am doing something really wrong here:

    Nodes    |    Current Node    |     Objective Bounds      |     Work
Expl Unexpl |  Obj  Depth IntInf | Incumbent    BestBd   Gap | It/Node Time

     0     0 26550.5890    0 1950   -0.00000 26550.5890     -      -    4s
     0     0 18754.1028    0 1854   -0.00000 18754.1028     -      -   12s
H    0     0                     239.5974000 18754.1028  7727%     -   13s
     0     0 17391.4529    0 1920  239.59740 17391.4529  7159%     -   16s
H    0     0                     244.2684000 17391.4529  7020%     -   16s
     0     0 17153.8979    0 1875  244.26840 17153.8979  6923%     -   18s
H    0     0                     244.8774000 17153.8979  6905%     -   18s
     0     0 17011.2374    0 1906  244.87740 17011.2374  6847%     -   19s
     0     0 16943.7668    0 1913  244.87740 16943.7668  6819%     -   20s
H    0     0                     245.8274000 16943.7668  6793%     -   21s
     0     0 16890.7751    0 1874  245.82740 16890.7751  6771%     -   21s
     0     0 16857.7210    0 1923  245.82740 16857.7210  6758%     -   22s
     0     0 16833.9578    0 1932  245.82740 16833.9578  6748%     -   23s
     0     0 16815.9366    0 1930  245.82740 16815.9366  6741%     -   24s
     0     0 16806.3253    0 1941  245.82740 16806.3253  6737%     -   25s
     0     0 16806.3253    0 1941  245.82740 16806.3253  6737%     -   26s
H    0     0                     248.0054000 16806.3253  6677%     -   37s
     0     0 16806.3253    0 1889  248.00540 16806.3253  6677%     -   43s
     0     0 16732.2340    0 1768  248.00540 16732.2340  6647%     -   56s
H    0     0                     249.7224000 16732.2340  6600%     -   57s
     0     0 16612.6719    0 1891  249.72240 16612.6719  6552%     -   58s
     0     0 16584.7054    0 1876  249.72240 16584.7054  6541%     -   59s

The gaps between the MIP bounds are larger than I have ever seen and don’t seem to get close anytime soon.

1 comment:

  1. That kind of situation happened to me when using models with penalty terms such as large backlog costs. Which the relaxation solution can eliminate, but the integer solution does not, which created large gaps between objective functions.

    ReplyDelete