Stopping the program if best objective has not changed after a while

Answered

Parisa Keshavarz

July 24, 2020 13:44

• 
• 
• 

Hello,

I have a MIP that takes a long time to solve and most of the time is spent on proving the optimal answer in the branch and bound algorithm. I though about setting a time limit, but in the following paper in the section 2.3. they mention that they tried terminating the program if the best objective has remained the same for some time. I wanted to know if such a thing is possible and how should I do it?

https://link.springer.com/article/10.1007/s10994-017-5633-9

Or what other parameters can I change or play with to make my program speed up. Note that my starting gap is over 1000% and it would be great if I could change that as well.

 

0

• Official comment

  

  Simranjit Kaur

  • Gurobi Staff

  December 06, 2024 00:43

  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?.
• 

  Eli Towle

  • Gurobi Staff

  July 24, 2020 22:43 Edited

  Yes, you can do this with a custom callback. Here is an example of a callback in Python that terminates the solve if the best solution has not improved for 20 seconds:

  import gurobipy as gp
  from gurobipy import GRB
  import time
  
  
  def cb(model, where):
      if where == GRB.Callback.MIPNODE:
          # Get model objective
          obj = model.cbGet(GRB.Callback.MIPNODE_OBJBST)
  
          # Has objective changed?
          if abs(obj - model._cur_obj) > 1e-8:
              # If so, update incumbent and time
              model._cur_obj = obj
              model._time = time.time()
  
      # Terminate if objective has not improved in 20s
      if time.time() - model._time > 20:
          model.terminate()
  
  
  m = gp.Model()
  
  # Build model here
  
  # Last updated objective and time
  m._cur_obj = float('inf')
  m._time = time.time()
  
  m.optimize(callback=cb)

  To solve the model faster, you could try setting the MIPFocus parameter (to 2 or 3, based on your description).

  1
• 

  Parisa Keshavarz

  July 25, 2020 06:32

  • 
  • 
  • 

  Dear Eli,

  Thank you so much for the quick response. I'm actually using MATLAB, is this also possible in MATLAB or any other method? I read that callback is not supported in MATLAB and I'm having difficulty for the speed. Or maybe is there a way to export the code to python without having to rewrite it?

  Thank you

  0
• 

  Eli Towle

  • Gurobi Staff

  July 28, 2020 00:15 Edited

  Ah, I looked at your previous posts and saw that you had used Python in the past. No, there is unfortunately no way to implement this in MATLAB, nor is there any automatic code generation tool to export your MATLAB/Gurobi code to another language.

  However, moving the model from the MATLAB API to the Python API isn't too hard. You can use the gurobi_write() MATLAB function to export your model to disk as an MPS file. Then, in Python, you can read that model file with the read() function. I.e., instead of constructing a new model with

  m = gp.Model()

  you read in the exported model file with

  m = gp.read('model.mps')

  Leave the rest of the code as-is and run the Python file.

  1
• 

  Parisa Keshavarz

  July 28, 2020 06:28

  • 
  • 
  • 

  Thank you so much Eli.

  0

Post is closed for comments.