About the calculation speed

Answered

Yanbo Fan

June 30, 2021 10:20

• 
• 
• 

Hi,

I'm solving a maximum concave problem. It's calculation speed is very slow. This is the LP file of my model.

When I simplify the model, the solution model contains the part about the decision variables, and the optimal solution can be obtained after four iterations. This is the LP file for the simplified model.

There is no difference between the two models except the coefficients of decision variables in the objective function. Why are their computing speeds so different? 

 

 

0

• Official comment

  

  Simranjit Kaur

  • Gurobi Staff

  December 06, 2024 00:29

  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?.
• 

  Jaromił Najman

  • Gurobi Staff

  June 30, 2021 12:56 Edited

  Hi,

  The coefficient values in the first model are way below Gurobi's tolerances. This then very likely leads to numerical difficulties such as decisions made based on round-off errors, (almost) wrong cuts, too many / too few cuts, and more... We recommend to keep the coefficient range as small as possible, especially for quadratic problems. You can learn more about tackling numerics in our Guidelines for Numerical Issues.

  Best regards,
  Jaromił

  0
• 

  Yanbo Fan

  June 30, 2021 13:28

  • 
  • 
  • 

  Hi Jaromił, 

  Thank you very much for your answer. Because of your answer, I already know the reason of the problem. Do I have any other methods besides manual control of the calculation accuracy in the calculation process?

  Best regards,
  Yanbo

   

  0
• 

  Jaromił Najman

  • Gurobi Staff

  June 30, 2021 14:06

  Hi Yanbo,

  Do I have any other methods besides manual control of the calculation accuracy in the calculation process?

  It is best to control the scaling of coefficient manually as far as possible. Obviously, this is not always possible. In this case, you might want to experiment with the NumericFocus parameter, which controls the effort put into numerical accuracy during algorithmic computations.

  Best regards,
  Jaromił

  0
• 

  Yanbo Fan

  July 01, 2021 06:16

  • 
  • 
  • 

  Hi Jaromił,

  Thank you again for your help.

  Best regards,

  Yanbo

   

  0
• 

  Yanbo Fan

  July 01, 2021 06:59

  • 
  • 
  • 

  Hi Jaromił,

  I'm sorry to trouble you again. I used numericfocus parameter, but the problem of slow solving speed has not been improved. Solver shows parameter numericfocus unchanged. 

  

  Is it the wrong way I use it?

  

   

  0
• 

  Jaromił Najman

  • Gurobi Staff

  July 01, 2021 07:53 Edited

  Hi Yanbo,

  You have set the NumericFocus parameter to 0, which is the default value for this parameter. Thus, Gurobi reports "unchanged". You can see all available values and the default value for the parameter in its documentation and reported by Gurobi as

  Parameter NumericFocus unchanged
      Value: 0  Min: 0  Max: 3  Default: 0

  So in your case you should try something like

  model.setParam("NumericFocus", 3) # or 1 or 2

  The higher you set the value of the parameter the more effort does Gurobi spend to tackle numerical issues. Please note that just setting the parameter may not be enough to help with severe numerical issues. Additionally, increasing the NumericFocus usually results in a performance decrease. Thus, it is always best to improve the numerics manually before experimenting with parameters.

  Best regards,
  Jaromił

  0

Post is closed for comments.