Overhead of adding lazy constraints via callback in Python

回答済み

Mike Lang

2022年04月24日 08:18

• 
• 
• 

Dear all, 

I am adding cuts to a MILP via lazy constraints with

model.cbLazy()

in a callback routine at

where == GRB.Callback.MIPSOL

 

I am looking at the time it takes to add the lazy constraints via the callback to identify levers for reducing runtime. I see that about 80% of the time in the callback is spent on adding the lazy constraint by calling 

 model.cbLazy()

, which equates to up to 10 seconds in some instances.

Is there information available on the overhead (in terms of runtime) of adding lazy constraints? 

Thank you!

0

• 

  Jaromił Najman

  • Gurobi Staff

  2022年04月25日 08:39

  Hi Mike,

  Could you please share a minimal reproducible example? Or at least one such hard coded lazy constraint which takes a long time to be added to the model? In particular, it is of interest how dense and numerically stable is your lazy constraint.

  Best regards, 
  Jaromił

  0
• 

  Mike Lang

  2022年04月26日 21:08

  • 
  • 
  • 

  Hi Jaromił, 

  thank you for the answer. Can I send the example to you somehow? Even the reduced version is a bit large to post it here. 

  Thank you!

  Mike

  0
• 

  Jaromił Najman

  • Gurobi Staff

  2022年04月28日 18:35

  The issue has been resolved internally. The problem was an inefficient usage of Python data structures. The Knowledge Base article How do I improve the time to build my model? holds many useful tips.

  0

サインインしてコメントを残してください。