Dnia 13 kwietnia 2016 16:26 Andrew Makhorin <[email protected]> napisał(a): I'm working with legacy code that uses GLPK. Here is the version number from the source code: #define GLP_MAJOR_VERSION 4 #define GLP_MINOR_VERSION 51 The legacy code solves many linear programming problems and sometimes the solver gets into an infinite loop. Below is part of the output from the infinite loop and it seems clear that this it a problem with numerical instability. My question is how can I deal with this: 1) Is there a newer version of GLPK that I should use where this problem would not occur? Yes. Please try the most recent version of glpk which is 4.60. It has improvements to avoid numerical instability and cycling. See lists.gnu.org lists.gnu.org . Would the new version always return some result without cycling or does it just cycle less often? 2) Can I do something to avoid the infinite loop with the version I'm using. This would be much preferred as the legacy code is undocumented. It would be best if I could eventually get some error from GLPK (and ignore this problem instance) rather it breaking the whole system. 3) I was considering running GLPK in an external process and killing it if no answer is produced in some time, yet this is laborious and would introduce problems with efficiency as the input for the process would have to be serialized and then parsed.
_______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
