1

I am solving QP problem with CVXPY and CPLEX for different number of variables. It worked fine for smaller number of variables but returned an error when I increased the number of variables.

image 1

image 2

Need some helps. Thank you.

Zhubei Federer
  • 1,274
  • 2
  • 10
  • 27
최시현
  • 21
  • 3

2 Answers2

1

In the log output you can see the message:

Barrier cannot determine infeasibility.

This indicates numerical issues. Also the error message from CVXPY in the end suggests that there may be numerical issues with your model/data.

You have to double-check your data. Maybe enable the data check parameter and the numerical emphasis parameter (I don't know how to set these things in CVXPY).

Another thing you can do is to export your model to an SAV (or LP) file with CVXPY, load that file into the CPLEX interactive optimizer and then do display prob stats. If that shows a large range of numbers then numerical issues are sort of confirmed. You could also solve the problem in the interactive optimizer and then look closely at the solution status returned there.

In case you want further help, you will have to update your question with

  • the output in case you set verbose to true
  • the exported model or at least the output of disp prob stats
Daniel Junglas
  • 5,830
  • 1
  • 5
  • 22
  • See [this](https://stackoverflow.com/a/51773264/1718477) stackoverflow answer for the details on setting CPLEX parameters in CVXPY. Setting CPLEX parameters and exporting the model is also described in the official CVXPY documentation [here](https://www.cvxpy.org/tutorial/advanced/index.html#setting-solver-options). – rkersh Jul 10 '19 at 13:20
  • Thanks a lot! I'll try it. – 최시현 Jul 11 '19 at 08:11
0

Although usually these prints mean that there is something wrong with your problem specification, this can sometimes rarely be a derp on the library side and be fixed with setting max_iters to a higher value.

ho4upivo
  • 28
  • 2