Topic
  • 1 reply
  • Latest Post - ‏2013-05-01T16:10:28Z by sval
sval
sval
2 Posts

Pinned topic Gap tolerance

‏2013-04-30T15:18:21Z |

Hi,

I'm working on a strange behavior and cannot figure what I should do.

I have a simple model that I run in OPL.

The last line of the log shows (either by running the model in the interface or by the command line; see attached file for the full logs ) :

*     0+    0                      1479369.9183  1482022.4099      995    0.18%

The gap is still significant, and I never succeed to close the gap. The best gap reached was 0.06. I played with different parameters to reach this value. Independently of the parameters chosen, when I try modifying my model inputs, the gaps at the end of each optimization are completely unpredictable, ranging between [ 0.06, 0.9 ].

Furthermore, when I run my model with usual inputs (same inputs that I used to log the optimization with OPL)  in CPLEX, the CPLEX log shows :

*  1923+  391                      1481545.5540  1481704.9239    17624    0.01%

I used a lp file generate from my OPL model in order to use CPLEX. The gap obtained with CPLEX is the gap that I will like to obtain in OPL. Both OPL and CPLEX use default parameters (I checked that default parameters are the same in OPL and in CPLEX), but the behavior seems pretty different. I don't know for what reason that happens, but the gap parameters in OPL are completely ignored when running my model.

Before testing in CPLEX, I was thinking that my model may have numerical instability. I improve my model without success in regard to the gap in OPL. Then, I found that CPLEX succeeds to solve the problem without trouble. Among all models I created so far, it's the only one having that behavior. The only significant difference between this model and my other models is that this model uses an Access files for importing input data.

Any ideas ?

Thanks,

Steve.

Attachments

  • sval
    sval
    2 Posts
    ACCEPTED ANSWER

    Re: Gap tolerance

    ‏2013-05-01T16:10:28Z  

    I finally find the problem. Like I thought, this is something pretty basic.

    Somewhere in the process, I copied/pasted some code from another model that was not mine. I didn't notice that there was some OPL script code that overwrite the parameter of cplex in this portion of model. Script code aren't exported, that explains why I got the correct behavior in CPLEX.

    Steve.

  • sval
    sval
    2 Posts

    Re: Gap tolerance

    ‏2013-05-01T16:10:28Z  

    I finally find the problem. Like I thought, this is something pretty basic.

    Somewhere in the process, I copied/pasted some code from another model that was not mine. I didn't notice that there was some OPL script code that overwrite the parameter of cplex in this portion of model. Script code aren't exported, that explains why I got the correct behavior in CPLEX.

    Steve.