Change solution strategy to standard (ImproveStartGap,ImproveStartNodes,ImproveStartTime)

Answered

Ilya Tyuryukanov

• April 20, 2022 21:28

I have a model, for which the standard solution strategy of the Gurobi solver works pretty fine (may be the best choice at all). But for some instances it helps a lot to focus on improving the bound (MIPFOCUS=3) at least for a short time in the beginning.

So I found these nice parameters ImproveStartGap, ImproveStartNodes, ImproveStartTime quite interesting, because they allow to focus on improving the bound for some time in the beginning and then switch to another strategy. Unfortunately, the strategy to which they switch involves lots of heuristics (seems similar to switching from MIPFOCUS=3 to MIPFOCUS=1).

I wonder if it is possible to choose another strategy to switch to after the conditions of ImproveStartGap, ImproveStartNodes, ImproveStartTime are fulfiled? I personaly would like to switch to the "default" solution strategy (not to Heuristics=0.5 RINS=10).

0

• 

  Maliheh Aramon

  Gurobi Staff

  • April 21, 2022 01:51

  Hi Ilya, 

  The parameters ImproveStartGap, ImproveStartNodes, and ImproveStartTime trigger solution improvement strategies where the Gurobi Optimizer only focuses on improving the incumbent to find a better feasible solution and essentially gives up on improving the best bound. For example, with the setting ImproveStartTime = 500, the Gurobi Optimizer starts with the default setting and after 500 seconds, it then just focuses on improving the incumbent. 

  I guess you would like to start with a parameter setting that does focus on improving the best bound (MIPFocus=3, for example) and then after reaching a certain gap, node counts, and/or time, switch back to the default setting. To do so, you can start the optimization with setting MIPFocus=3 and check the gap, node count, and/or time in the MIP callback. You can then terminate the optimization in the callback as soon as reaching the specified limits, reset the parameters to their default values, and continue the optimization. Please check out our article on How do I change parameters in a callback?.

   Best regards,

  Maliheh

  1
• 

  Ilya Tyuryukanov

  • May 05, 2022 00:17

  

  

  

  Thanks, this has worked. Very much appreciated.

  0

Please sign in to leave a comment.