Programmatically determining warnings, e.g. large RHS

Answered

Frederik Schadd

• October 11, 2023 08:28

Hello,

I'm currently facing a model issue that's causing the following warning:

Warning: Model contains large rhs

While I've figured out the actual model issue already, I'm wondering if there is a programmatic way of determining whether your model has this or any other similar issue prior to actually solving (i.e. large coefficients). I'd like to add a sanity check into our build process such that we can immediately detect and flag these issues whenever they pop up.

Related to

• Numerics
• Log Files

0

• 

  Matthias Miltenberger

  Gurobi Staff

  • October 11, 2023 08:52

  Hi Frederik,

  Apart from checking your data before or while building the model, you can also run a quick sanity check before optimizing it.

  This will print the model statistics:

  model.update()
  model.printStats()

  Please note that you must process pending model changes by calling update() before displaying the statistics.

  I hope that helps!

  Cheers,
  Matthias

  0
• 

  Frederik Schadd

  • October 11, 2023 09:33

  

  

  

  Hi Matthias,

  thanks for the response. This approach would print the stats to the stdout which I would then have to read back in again. Is there a more direct of doing this? Also, what are the thresholds that gurobi applies to decide whether to throw a warning? 

  0
• 

  Marika Karbstein

  Gurobi Staff

  • October 11, 2023 14:19

   You could also check the Model Attributes, for example, MaxRHS. 
  The threshold for the warning is 1e+09 for rhs, bounds, and obj coefficients. It is 1e+08 for matrix coefficients.
  But we recommend to even have smaller values, see Recommended ranges for variables and constraints and Advanced user scaling.

  1

Please sign in to leave a comment.