Originally posted by: MarekGrzes
I am trying to solve a fairly large MIP formulated in OPL. When I run the solver with 2 hour time limit, I obtain the following:
<<< generate
Presolve has eliminated 33928 rows and 1 columns...
Presolve has eliminated 33928 rows and 1 columns...
Tried aggregator 1 time.
Presolve has eliminated 622860 rows and 1 columns...
Presolve has improved bounds 1179648 times...
MIP Presolve eliminated 622860 rows and 1 columns.
MIP Presolve modified 14556 coefficients.
Reduced MIP has 2364181 rows, 593664 columns, and 6315528 nonzeros.
Reduced MIP has 2304 binaries, 0 generals, 0 SOSs, and 0 indicators.
Elapsed time 528.24 sec. for 0% of probing.
Elapsed time 1051.65 sec. for 0% of probing.
Elapsed time 1407.03 sec. for 0% of probing.
Elapsed time 1754.88 sec. for 0% of probing.
Elapsed time 2100.45 sec. for 0% of probing.
Elapsed time 2451.27 sec. for 0% of probing.
Elapsed time 2977.12 sec. for 0% of probing.
Elapsed time 2987.29 sec. for 0% of probing.
Elapsed time 2997.30 sec. for 1% of probing.
Elapsed time 3007.73 sec. for 1% of probing.
Elapsed time 3546.77 sec. for 1% of probing.
Elapsed time 4065.64 sec. for 1% of probing.
Elapsed time 4575.43 sec. for 2% of probing.
Elapsed time 4927.11 sec. for 2% of probing.
Elapsed time 5274.36 sec. for 2% of probing.
Elapsed time 5610.96 sec. for 2% of probing.
Elapsed time 5938.82 sec. for 2% of probing.
Elapsed time 6277.36 sec. for 2% of probing.
Elapsed time 6616.17 sec. for 2% of probing.
Elapsed time 6626.37 sec. for 2% of probing.
Elapsed time 6998.87 sec. for 3% of probing.
Presolve time = 7395.20 sec.
<<< solve
<<< no solution
which shows that within 2 hours time limit the algorithm could not even start B&C. My model is large so this is probably not a surprise. I have one question related to OPL. In my current OPL model, I do not use tuples. There is one large matrix in my model which is very sparse, and I encode it as an OPL matrix (I do not use tuples). Since my matrix is sparse (it contains a lot of zeros), I assume that presolver will still use a sparse representation internally (though foreach and sum in my OPL model are over all dimensions of that matrix). Do you think that I could gain anything, encoding that matrix using tuples? My decision variables are in another large matrix, so that matrix cannot be represened using tuples because it contains decision variables. If anybody has some thoughts or experience, please let me know.
Is my above model large enough to expect CPLEX taking so much time in the preprocessing stage?
Cheers,
Marek
#DecisionOptimization#OPLusingCPLEXOptimizer