Re: Re[Help-glpk] garding speeding up MIPs by providing Initial solutions
Thanks for your reply. Please see my reply in the lines. I've tested your model. In fact, it is not hard for glpk 4.37 and can be solved in several minutes. However, it is hard enough to be solved faster. The model looks like a pseudo-boolean problem (more precisely, like a satisfiability problem), so you might try to use specific solvers. Why do you need to solve the mip many times? Is it a subproblem? Do you really need optimal solution, or a suboptimal solution would be sufficient to you? Because I need to change the constraints according to the solution and solve the MIP again. For a MIP, I need an optimal solution. For my big problem, of course , it is suboptimal. If I form my big problem into a mip, then it will be huge and takes forever to run. I think that in many cases using a complete description is more reasonable. If, for example, the complete mip is huge because it has too many rows (constraints), you could use row generation technique. Andrew Makhorin ___ Help-glpk mailing list Help-glpk@gnu.org http://lists.gnu.org/mailman/listinfo/help-glpk
Re: [Help-glpk] problem with glpsol
Here is the cplex lp file and the mps file. Thanks. Wrong results appear, because the problem is badly scaled. This is caused by two tiny constraint coefficients: C036 R033 -5.16198E-27 R034 -0.079105507 C036 R039 -0.092756701 R040 -5.51167E-27 To avoid the error you should either disable the geometric mean scaling (which is enabled by default) or, that is much better, replace all tiny constraint coefficients by exact zeros, for example, before passing rows/columns to glp_set_mat_row/glp_set_mat_col. BTW, I could not reproduce the error. Andrew Makhorin ___ Help-glpk mailing list Help-glpk@gnu.org http://lists.gnu.org/mailman/listinfo/help-glpk
[Help-glpk] Obtain the variables values in a GLPK 4.8 Java Interface
Hello, I am currently using GLPK with the Java interface developped and made available in the following website: http://bjoern.dapnet.de/glpk/ When I solve a LP and I want to retrieve the values of the solution (using getColPrim() and getColDual(), for example), these values (they are of type double) are always rounded to the lower integer. Yet, the routine printSol(), that prints the solution into a file, doesn´t approximate these values. The objective value (accessed by getObjVal()) isn´t approximated either. Is there some parameter to customize this rounding? Or do I have to contact the designer of the Java interface to solve my problem? I am sorry I cannot provide a piece of code quickly (mine is rather complex), but I hope you won´t need it to answer my question. Sylvain ___ Help-glpk mailing list Help-glpk@gnu.org http://lists.gnu.org/mailman/listinfo/help-glpk