Runtime of Solver

Katharina König

• June 25, 2019 14:18

Hello, 

I am new to Gurobi and not really advanced in the topic of optimization too.

I created a MIP model in Python and I am using the gurobi solver to solve it. 

As i did not have the real data in the beginning I used synthetic timeseries for building the model and it worked fine with it. By now I am using the real data. The only difference is that some of the real values are much smaller than the synthetic ones.

I changed nothing but the values but the runtime of the model increased extremely. Before I reached a 1 % gap after 50 seconds. Now it needs around 2 hours for the same. 

Does anyone have an idea what the problem is? Here is the header of the gurobi solving:

 

And is there some rule of thumb to conclude the expected time needed for the solving from the number of variables?

In the end I am planing on solving a problem with 192720 continous and 17520 binary variables. It would be great to know in advance if that is even possible in a reasonable amount of time. 

 

Thanks a lot for any help!

Rina

0

• 

  Sonja Mars

  Gurobi Staff

  • August 02, 2019 08:55

  Hi,

  Can you share complete log files for these runs? Without them it is tricky to say, why it needed 50 seconds and now needs 2 hours.

  Also please note, there is no way of predicting the run time of a MIP just from the size. It depends on so many factors. I have seen very large models that solve in a few minutes but I have also seen very small models that takes days to solve.

  Best regards,

    Sonja

  0
• 

  Katharina König

  • August 04, 2019 07:03

  

  

  

  Hey Sonja,  

  thanks for your answer. I found some workarounds to make the solver faster. 

  But I have another question now and I hope you can help me with it. 

  Since yesterday I get shown the barrier log even though I don´t know what why. 

  What confuses me more ist that when I print my model statistics after finding the solution using

  m.printQuality()

  I get :

  Solution quality statistics for model 15 energy system_copy :
  Maximum violation (unscaled/scaled):
  Bound : 0.00000000e+00 / 0.00000000e+00
  Constraint : 7.26562500e-07 / 7.26562500e-07 (Speicher_min[278])
  Dual : 3.55271368e-15 / 3.55271368e-15 (EHU[91])

  My model is named '15 energy system'. Why is it showing me the statistics of a copy of my model?

  0
• 

  Katharina König

  • August 04, 2019 07:12

  

  

  

  Before I always had the MIP logging.

  0
• 

  Sonja Mars

  Gurobi Staff

  • August 22, 2019 06:52

  Can you share the whole log? Do you have a copy of the model object around or is m the only Gurobi model?

   

  0

Please sign in to leave a comment.