Does ScaleFlag not work?

Answered

Bodo Lipp

• March 14, 2024 17:57
• Edited

I am wondering if the ScaleFlag = 2 is doing anything in my case. Below is part of the logfile. There you can see that Gurobi reports large values/ range in the optimization problem. The coefficient statistic looks exactly the same as when I do not use the ScaleFlag option. So, was there any scaling applied in my case? And if so, why do I see the coefficient statistic before scaling and not after?

Set parameter Method to value 2
Set parameter ScaleFlag to value 2
Set parameter BarConvTol to value 1e-05
Set parameter Crossover to value 0
Set parameter LogFile to value ...
Set parameter Threads to value 15
Gurobi Optimizer version 10.0.0 build v10.0.0rc2 (linux64)

...
Thread count: 32 physical cores, 32 logical processors, using up to 15 threads

Optimize a model with 190952105 rows, 145453003 columns and 350948499 nonzeros
Model fingerprint: 0x2b2a7815
Coefficient statistics:
  Matrix range     [1e-08, 1e+06]
  Objective range  [1e-03, 2e+05]
  Bounds range     [3e-12, 4e+10]
  RHS range        [2e-01, 6e+08]
Warning: Model contains large matrix coefficient range
Warning: Model contains large bounds
         Consider reformulating model or setting NumericFocus parameter
         to avoid numerical issues.

Related to

• Scaling

0

• 

  Michel Soares

  • March 14, 2024 19:08

  

  

  The coefficient statistics, by default, are always shown for the original model, before any scaling.

  Please try running this to see if the changes after presolving, which should include your scaling:

    p = model.presolve()
    p.printStats()

   

  0

Please sign in to leave a comment.