Advice on Cuts and Reducing Solution Time

進行中

William ONeill

2020年11月16日 22:02

• 
• 

Asking for advice and a sanity check.

I was wondering if someone could possibly give me some advice on cuts. I'm wondering if I should avoid or prioritize certain cuts given my issue.

 

I have a moderate to large MILP with some general constraints/big-M constraints. I've spent a lot of time systematically testing constraints for improvement and testing GUROBI's settings. what I have now seems to reach feasibility and reach a low gap% rather quickly. 

 

My largest problem, and the one that is the road-block of my dissertation, gets close to convergence but not quite there. It seems to get close to the threshold gap but takes days to close. See the attached chart and log. I might be able to call that converged and just set the MIP gap to that and call it a day but I'm wondering if anyone has any advice or tips. 

My settings:

-MIP gap: 1E-3
-Cuts: 1(found 1 gave me better performance)

-MIP Focus 3

-VarBranch: Pseudo Reduced Cost Branching

-32 thread 2.6g node

Example cuts from a case that was almost solved but stalled. I've tried changing a few of these individually as well as changing the global cuts. Does anything pop out?

Cutting planes:

  Learned: 1
  Gomory: 47
  Cover: 68
  Implied bound: 19
  Projected implied bound: 7
  Clique: 4
  MIR: 205
  StrongCG: 3
  Flow cover: 242
  Inf proof: 3
  Zero half: 11
  RLT: 18
  Relax-and-lift: 19

 

Full log here https://drive.google.com/file/d/1WZuA2GlvcwLLxAnpzgi9x39N8TVDOIyg/view?usp=sharing

1

• 正式なコメント

  

  Simranjit Kaur

  • Gurobi Staff

  2025年06月23日 03:04 編集済み

  This post is more than three years old. Some information may not be up to date. For current information, please check the Gurobi Documentation or Knowledge Base. If you need more help, please create a new post in the community forum. Or why not try our AI Gurobot?.
• 

  Matthias Miltenberger

  • Gurobi Staff

  2020年11月17日 11:28

  Hi William,

  I just edited your post. Please don't create such long posts - especially if you already include a link to the log. The linked log is not complete by the way: the final statistics are missing.

  Concerning your actual question: As you probably know, MILPs can be extremely hard to solve and proving optimality or at least a certain gap is often the hardest part. There might not be a simple solution to your problem. I can only recommend testing a few parameter sets to see how far you can push the solver.

  Cheers,
  Matthias

  0

投稿コメントは受け付けていません。