You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am currently working on a real class of scheduling problem, let me call Molding process.
In short: the problem is categorized as $(P_{m},s_{1} | r_{j}, due_{j}, S_{i,j}, pred, cap | C_{max}/JIT)$. As the complexity of the problem is high and standard MIP solvers, like Gurobi and Cplex, cannot solve the problem in a reasonable amount of time, I presumably need to use a decomposition scheme.
I have tried to test some of the formulations to see how their modeling structure can perform better in the MIP solvers. Now, I have some questions and I was wondering if,
In the GCG doc, you mentioned that the problem structure can either be exploited by software or the user. Would you please, elaborate more on the second form?
In some instances, when the software detects the number of ways the problem can be decomposed, how does it really decide which one would be selected? (The one with the max $maxWhiteScore$?)
If the solver decides based on $maxWhiteScore$, in many instances, the problem with a lower score can be solved faster than the one with an upper score!!!
Please, see the following log: 0 Decomp scores: d.classicScore=-1.0000, d.borderAreaScore=-1.0000, d.maxWhiteScore=0.0000, d.maxForWhiteScore=0.0000
and corresponding matrix and the solving process is:
while when I choose the number with a $maxWhiteScore$ of around 0.6, it takes so longer time to solve or even it stops at time limit criteria! (In many cases, cannot find any feasible solution). Would you say, is it normal behavior or I am missing something?
I assume that the performance of the GCG in the worst case would be the same as SCIP. While SCIP can solve the above problem almost in 209 seconds, GCG cannot reach a feasible solution also in many cases it can decompose the problem. Is my intuition correct? If not, please correct me.
All the best
Regards
The text was updated successfully, but these errors were encountered:
It is possible for the user to define a decomposition in pygcgopt if he has an idea of the structure.
Our default score is the set partitioning foreseeing max white score (spfwh). (see https://gcg.or.rwth-aachen.de/doc-3.5.0/detection-scores.html)
3.-4. All scores are heuristics and topic of current research. GCG is mostly only better than SCIP on structured models.
Dear support team,
I am currently working on a real class of scheduling problem, let me call Molding process.$(P_{m},s_{1} | r_{j}, due_{j}, S_{i,j}, pred, cap | C_{max}/JIT)$ . As the complexity of the problem is high and standard MIP solvers, like Gurobi and Cplex, cannot solve the problem in a reasonable amount of time, I presumably need to use a decomposition scheme.
In short: the problem is categorized as
I have tried to test some of the formulations to see how their modeling structure can perform better in the MIP solvers. Now, I have some questions and I was wondering if,
In the GCG doc, you mentioned that the problem structure can either be exploited by software or the user. Would you please, elaborate more on the second form?
In some instances, when the software detects the number of ways the problem can be decomposed, how does it really decide which one would be selected? (The one with the max$maxWhiteScore$ ?)
If the solver decides based on$maxWhiteScore$ , in many instances, the problem with a lower score can be solved faster than the one with an upper score!!!
Please, see the following log:
0 Decomp scores: d.classicScore=-1.0000, d.borderAreaScore=-1.0000, d.maxWhiteScore=0.0000, d.maxForWhiteScore=0.0000
and corresponding matrix and the solving process is:
while when I choose the number with a$maxWhiteScore$ of around 0.6, it takes so longer time to solve or even it stops at time limit criteria! (In many cases, cannot find any feasible solution). Would you say, is it normal behavior or I am missing something?
All the best
Regards
The text was updated successfully, but these errors were encountered: