-------- Forwarded Message -------- From: Alejandro Pascual <[email protected]> To: Andrew Makhorin <[email protected]> Cc: [email protected] Subject: Re: [Help-glpk] Is it possible to feed glpsol with a feasible initial solution to a MIP problem? Date: Sat, 7 Apr 2018 09:43:32 -0300
Oops sorry, I messed up versions 4.61 and 4.65 between the first and second execution of glpsol. Being consistent with v4.65 I get fewer warnings. Anyway, I can't get glpsol to use the feasible solutition fed with --ini. The integer optimization stage ends (after a generous time limit) with status INTEGER UNDEFINED in the output file (I attach only the log file of the second executiion because the output file is too big). I can't figure out what I am doing wrong. Regards, Alejandro 2018-04-06 21:19 GMT-03:00 Alejandro Pascual <[email protected]>: > Dear Andrew, > > Thanks for your quick reply, I'm still a beginner with GLPK and > haven't noticed the -w and --ini options. > > I've tried the --fpump option with no success (even after waiting for > days), and after your suggestion, the --proxy option with 5400 seconds > without success either. I attach the model and data files. Setting the > binary variables to one, glpsol can find a feasible solution. > > Now I'am trying the -w, --ini sequence of two consecutive glpsol > executions (-w to save the solution obtained with the binary variables > set to one, and --ini to feed this solution as a feasible solution to > the original MIP problem). After the LP relaxation, the second > execution is warning me: basis matrix is ill-conditioned (cond = > 3.27e+13). I guess the instance of the problem is just too hard, but I > am not sure if the integer optimization stage is actually using the > feasible solution I intended to provide. ¿How should I understand the > attached log file? > > Thank you very much for your help. > > Best regards, > Alejandro > > 2018-04-06 6:13 GMT-03:00 Andrew Makhorin <[email protected]>: >> On Thu, 2018-04-05 at 14:40 -0300, Alejandro Pascual wrote: >>> Hi, >>> >>> I am working on a MIP problem for which glpsol can't find a feasible >>> solution in a reasonable amount of time. However, fixing the values of >>> the integer variables (which are all binary) to one, I can get glpsol >>> to find an optimum solution rather quickly. Is it possible to feed >>> this solution as initial (suboptimal) solution of the original >>> problem? >> >> Yes. Save the mip solution found with 'glpsol ... -w FILENAME' and then >> specify it as initial one 'glpsol ... --ini FILENAME'. >> >> BTW, did you try using --fpump or --proxy to find a feasible solution? >> >>> >>> I have read in the help-list that this was not possible with MathProg >>> in 2007, but maybe now it is. If it is not, what would be the most >>> direct way to hack the code in order to do so? >>> >>> Regards, >>> Alejandro Pascual >>> >>> _______________________________________________ >>> Help-glpk mailing list >>> [email protected] >>> https://lists.gnu.org/mailman/listinfo/help-glpk >>> >> >> _______________________________________________ Help-glpk mailing list [email protected] https://lists.gnu.org/mailman/listinfo/help-glpk
