search

scipopt / PySCIPOpt

Python interface for the SCIP Optimization Suite
https://pyscipopt.readthedocs.io/en/latest/
MIT License
826 stars 255 forks source link

a question about solving speed of MILP problem #419

Closed Anyon123 closed 4 years ago

Anyon123 commented 4 years ago

I have a MILP problem, worte in an .lp file. When I used SCIP to solve it. There is a warning like "Deactivated symmetry handling, since SCIP was built without symmetry detector". The problem was solved after 900s later. But when I used Gurobi to solve it, it spent like 1 second. I wonder if the symmetry handling is so important that affect several hundred times of solving time? And what should I do to reduce the solving time by SCIP? Thank youu in advance

mattmilten commented 4 years ago

This might not even be only due to the missing symmetry. Gurobi is very well-tuned for performance on a very large set of instances. You may be able to achieve comparable performance with SCIP/PySCIPOpt when setting the correct parameters. I suggest you inspect the log files of the two solvers and try to figure out why SCIP takes so much longer. Solving MILPs is a difficult task and sometimes you just have to wait a bit for the solution. Another approach could be to relax your termination criteria, for example a larger MIP gap might decrease the solving time a lot.

Anyon123 commented 4 years ago

Thank you very much dear mattmilten, you helps a lot