Decision Optimization

Decision Optimization

Delivers prescriptive analytics capabilities and decision intelligence to improve decision-making.


#Analytics
#DecisionOptimization
#DecisionOptimization
 View Only

  • 1.  Branching priority

    Posted 10/13/17 08:25 AM

    Originally posted by: yanzhiping


    In a previous post, @EdKlotz suggested branching priorities on MIP.

    https://www.ibm.com/developerworks/community/forums/html/topic?id=602a1d9d-dbf6-4b19-ba60-720e40cbb946&ps=25

    Based on your model description, your model has time periods, and it also has variables y that depend on variables x.   Both of these suggest that branching priorities might help performance.   Give higher priority to the earlier time periods, and within a given time period, give high priority to the independent variables x.

     

    But In GAMS/cplex, GAMS support team suggested that

    MIP solvers improved significantly over the last years and decades. Part of this improvement is for sure to make good branching decisions.
    It is impossible to forecast whether you can easily find a manual branching prioritization that beats CPLEX' default for your specific problem. What the manual says is that it might have been easier in the past but nowadays letting the solver decide when to branch on which variable works very good for most problems and it is difficult to come up with better manual branching priorities.

     

    Is it necessary to do branching priority or not?

     

    Thanks,


    #CPLEXOptimizers
    #DecisionOptimization


  • 2.  Re: Branching priority

    Posted 10/14/17 12:11 AM

    Originally posted by: EdKlotz


    Branching priorities are certainly not necessary.    The GAMS team is correct that default branching in commercial MIP solvers has improved significantly, so that branching priorities are not as important as they were say 20 years ago.   But, if you are looking for better performance, and you have tried various non default parameter settings, and your model has some clear priorities among the discrete variables, they are worth trying.


    #CPLEXOptimizers
    #DecisionOptimization


Related Content