0

I'm using SCIP with Python (via Jupyter Notebook) and solve an ILP. When I change the input numbers it is possible that the result values of the integer variables are not integer anymore. I've seen in a post (http://listserv.zib.de/pipermail/scip/2013-December/001748.html) that this seems to be expected behavior for a very small range of numbers (e-06 - e-09) which is fine by me.

But I have now an instance where this is happening at a larger scale. Number that should be 3.0 are 2.599999999999998 which scares me a bit since this is a case close to rounding down and far of the described e-06 tolerance.

Does anyone have an idea why the differences are that large or if there is anything I can do to make my models more robust.

Thanks for any hints and your support.

stahldonnerklinge
  • 1
  • 2
  • 1
    Such a deviation is of course not acceptable and SCIP should not return such values for integer variables. My guess is that something else is broken between your input data and this result. Could you please share your model/code? – mattmilten May 29 '18 at 12:36
  • Hi Mattmilten Thank you for your fast Reply. I'm trying to revert back to that version. I should have saved the state of the model, but did not. I will post when I established the concrete case again. Sorry for the inconcenience! – stahldonnerklinge May 30 '18 at 11:23
  • I could reconstruct the case. But here is my finding: The instance was infeasible! Since I was focused on my output (where I did not output the feasibility Status) I did not see this. Also I assumed that in case of infeasibility there would be something like NaN numbers. But that was my wrong assumption. Anyways SCIP runs to smooth and seems to be tested very I had doubt that there were something wrong in the lib. But getting back to people always help. ;-) Thanks for your support! – stahldonnerklinge May 31 '18 at 09:19

0 Answers0