TimeLimit of Root node and quickly proceed to Branch and bound

Awaiting user input

ce jekl

• March 14, 2022 14:31

How to limit the time spent in the root node, since I find after a long time in root node there's no new incumbent of my problem. I want the model to quickly proceed to the phase of branch and bound.

0

• 

  Jaromił Najman

  Gurobi Staff

  • March 16, 2022 10:07
  • Edited

  You should have a look at the parameters listed in our MIP Models documentation. In addition, you could add a snippet of your log output and maybe someone in the Community is able to observe something interesting.

  Best regards, 
  Jaromił

  0
• 

  ce jekl

  • March 16, 2022 10:00

  

  

  

  Thank you, I think I'm pretty familiar with the information on the website

  But my question is how can I quickly move from the root node to the descending nodes without spending lots of time in the root node since I know that there will be no solution in the root node from both experience and experiments.

  I try to set Heuristics=0.0001 and Cuts=0, but gurobi still spends lots of time in the root node.

  Best regards,

  cero

  0
• 

  Jaromił Najman

  Gurobi Staff

  • March 16, 2022 10:12

  You could try the No Relaxation heuristic to find a feasible point before the root node relaxation.

  Turning off Cuts and Heuristics makes sense to get through the root node quicker. You could try experimenting with the Presolve parameter to reduce the size of the presolved model. Is it possible that your model suffers from numerical issues?
  Could you post a log snippet to show the behavior? It is hard guessing what might help and what might not.

  Best regards, 
  Jaromił

  0
• 

  ce jekl

  • March 16, 2022 17:28
  • Edited

  

  

  

  Thanks! The problem magically disappears, but I have one more question.

  If I optimize a multi-objective model after setting the first obj and then add more objs and optimize, without changing any of the constraints. The model won't use the solution of first optimize as user MIP start since I find no "Load user MIP start" or "Use previous solve" in the log and of course there's no incumbent in the log.

  

  However, I write the solution to "a.sol" after the first optimize and read it back before the second optimize, the log displays "Load user mip start". and it has incumbents once entering the mip log. Do you have any idea why this situation happens?

  0
• 

  Jaromił Najman

  Gurobi Staff

  • March 17, 2022 11:25

  What you observe is currently intended behavior. After adding additional objectives, the data structure gets manipulated in a way that it cannot reuse a previously computed solution. This is circumvented when you write and read the solution point, because it takes a different route to get into the new multi-objective model. This is something, we will try to improve with the next release. As a workaround, you can either write and read a solution file or save the solution values in a list and set the variable's Start attribute.

  Best regards, 
  Jaromił

  0
• 

  ce jekl

  • March 21, 2022 03:46

  

  

  

   

  Gurobi spends lots of time in the root node, but this problem is nearly infeasible, so it cannot find a feasible solution quickly. Can I skip the root node?

  0
• 

  Jaromił Najman

  Gurobi Staff

  • March 21, 2022 08:01

  Unfortunately, it is not possible to complete skip  the root node. Looking at the Simplex iterations per node column ("It/Node") it looks like your model might have some numerical issues. Would it be possible for you to share this particular model as described in Posting to the Community Forum? Maybe there is some setting that might help in this case.

  Best regards, 
  Jaromił

  0

Please sign in to leave a comment.