Decision Optimization

 View Only
Expand all | Collapse all

(Bug?) Can't explain behavior of running Cplex on the same file with different amount of threads.

  • 1.  (Bug?) Can't explain behavior of running Cplex on the same file with different amount of threads.

    Posted Wed July 24, 2024 01:35 PM

    Hi,
    I encountered some weird behavior in CPLEX I think qualifies as bug.

    When I do:

    read problem.lp
    optimize

    the returned objective value is 3. Whereas when I do:

    read problem.lp
    set CPXPARAM_Threads 1
    optimize

    The returned objective value is 1. I tried playing around with CPXPARAM_MIP_Tolerances_MIPGap , but it had no impact on the observed behavior. Is there some reasonable explaination to this or is it simply a bug? If it is a bug, where should I report it?

    I am new to this community, so I am happy to receive any feedback or guidance. Thanks in advance for any guidance. 



    ------------------------------
    Pascal Lauer
    ------------------------------

    Attachment(s)

    log
    attempt_1_core.log   1 KB 1 version
    log
    attempt_8_core.log   1 KB 1 version
    lp
    problem.lp   8 KB 1 version


  • 2.  RE: (Bug?) Can't explain behavior of running Cplex on the same file with different amount of threads.

    Posted Thu July 25, 2024 05:11 AM

    Hi,

    if you use datacheck 2 you will get warning about large values.

    When the value of this parameter is set to level 2, CPX_DATACHECK_ASSIST, CPLEX turns on both data consistency checking and modeling assistance. At this level, CPLEX issues warnings at the start of the optimization about disproportionate values (too large, too small) in coefficients, bounds, and righthand sides (RHS).

    If you turn your big M to 10001 your model will work better.

    Can you try that ?



    ------------------------------
    [Alex] [Fleischer]
    [Data and AI Technical Sales]
    [IBM]
    ------------------------------



  • 3.  RE: (Bug?) Can't explain behavior of running Cplex on the same file with different amount of threads.

    Posted Thu July 25, 2024 06:28 AM

    Ohh, I see. Yes this works. Thank you very much. :)



    ------------------------------
    Pascal Lauer
    ------------------------------