MILP problem solving - Crossover
AnsweredDear Forum,
I am trying to solve an MIP optimization problem using Gurobi. I first tried to solve it using Method 2 (Barrier) and Crossover set to 0. The optimal objective is reached within 60 iterations in a reasonable amount of time, but then the following starts to happen, which I understand is part of the crossover:
-----------------------------------------------------
Set parameter Method to value 2
Set parameter Crossover to value 0
(...)
60 4.25413242e+10 4.25290609e+10 1.99e-03 9.02e-09 1.45e-01 34075s
Barrier solved model in 60 iterations and 34076.12 seconds (17385.85 work units)
Optimal objective 4.25413242e+10
Root crossover log...
8126993 variables added to crossover basis 34103s
25259695 DPushes remaining with DInf 0.0000000e+00 34187s
-----------------------------------------------------
--> It goes on like this for ages...
I do not understand why this is happening if I set crossover to 0. Any recommendations about what I should specify to solve this problem the fastest? I am now trying with Method=5 (primal & dual simplex), but this one takes much longer to reach the optimal objective.
Thank you in advance for your help!
Best regards,
Sebastián
-
Hello Sebastián,
The Method and Crossover parameters only affect the algorithm for the initial root relaxation of your MIP. What follows when solving a MIP is the Branch and Bound algorithm. The algorithm that solves node relaxations within the Branch and Bound algorithm is defined using the NodeMethod parameter; where Primal and Dual Simplex expect basic solutions from the previous node solves, which you do not have if you ran Barrier without Crossover.
To disable Crossover completely, you would have to also set NodeMethod=2 to enforce Barrier as the algorithm to solve all MIP node relaxations. Without knowing the details of your model, I see a risk that this parameter setting could harm overall MIP solution performance, although you saw speed-ups for the initial root relaxation. But feel free to try it out and let us know what you observed.
Best regards,
Lennart0 -
Hello Lennart,
Thank you very much for your answer, it was helpful for me to better understand the situation I was in.
Best regards,
Sebastián
0
Please sign in to leave a comment.
Comments
2 comments