Decision Optimization

 View Only
Expand all | Collapse all

Different behavior of cplex when use different time limit

  • 1.  Different behavior of cplex when use different time limit

    Posted Tue September 07, 2021 10:09 AM
    Hi everyone!

    I have seen that when I optimize a MIP using a time limit of 600s cplex can find the first feasible solution after 18 s. However, when I change the time limit to 3600s cplex find the first feasible solution after 180s. Why does it happen? and how can I avoid this? Comparing logfiles (attached), it seems to be that something different happens during the presolve.

    I'm using Cplex version 20.1 and the following parameters:

    cplexE.setParam(IloCplex::TiLim, dKSTime);// dKSTime
    cplexE.setParam(IloCplex::ParallelMode, CPX_PARALLEL_DETERMINISTIC);//set para 1
    cplexE.setParam(IloCplex::Threads, 1);//set threa 1
    cplexE.setParam(IloCplex::MIPEmphasis, CPX_MIPEMPHASIS_FEASIBILITY);

    I will appreciate your help to understand this behavior.

    Best,

    Diana.

    ------------------------------
    Diana Huerta
    ------------------------------

    #DecisionOptimization


  • 2.  RE: Different behavior of cplex when use different time limit

    Posted Tue September 07, 2021 10:38 AM
    Hello, can you provide a LP file (.lp or anonymised .alp file) to reproduce on our side?

    We checked with several models and don't see this behavior.
    It would help understand and explain it.

    ------------------------------
    Vincent Beraudier
    ------------------------------

    Original Message


  • 3.  RE: Different behavior of cplex when use different time limit

    Posted Tue October 05, 2021 02:53 PM
    Thank you, Vincent.

    Sorry for the late reply, this message arrived to spam.

    Currently I have done some changes to my algorithm and I have seen that the LPs are slightly different. First I solve an LP model to optimality to obtain the information about which variables to fix to zero in my MILP, and this LP provided a different combination of variables, very similar but different at the end (I'm not really sure if it should be like this, I will investigate more the causes). I made the same test considering all the variables in the MILP, then solve it using 2 different time limit and I only see some changes in the computing time, so I need to work on it to ensure by my side if this is a normal behavior. For now, I think it is better to close this question, once I know more about it I will write a new dicussion if need it.

    Thanks Vincent.

    Diana.

    ------------------------------
    Diana Huerta
    ------------------------------

    Original Message