Amazing results with ODH|CPLEX
Download as PPTX, PDF0 likes194 views
The document presents results from using Optimization Direct's ODH|CPLEX on a collection of 286 models, revealing optimal solutions for 9 models and better solutions for 39% of the cases within 2 hours. It highlights performance metrics such as feasible solutions, optimality proofs, and significant gaps found compared to best-known solutions, emphasizing the effectiveness of parallel processing. The findings indicate increasing customer demand for solving larger models, with ODH|CPLEX demonstrating capacity to yield high-quality results even on smaller scales.
Amazing results with ODH|CPLEX
- 1. Optimization Direct Inc. October 2019 Amazing results with ODH|CPLEX
- 2. Miplib Open-v7 Models • Public collection of 286 models to which an optimal solution has not been proven • 257 models are known to have a feasible solution • No solution found to 29 models • Tried ODH|CPLEX on this set • Proves optimality on 9 models • Finds better solutions than the ‘best known’ in 2 hours to 101 (39%) of them with 8 threads (43% with 16 threads) • Finds solutions to 5 models where no solution found before
- 3. Better Solutions: 8 Threads, 2 hrs • Optimality on 6 models (2 absolute) • Better solutions on 39%, matched on 19% • ‘Top Ten’: Model Best known ODH Soln Diff Gap Time rmine25 -2553.479 -10831.51 324.2% 240% 2792 rmine21 -2645.005 -10500.37 297.0% 1.70% 2613 mining 531134254.8 -793124325 249.3% 16.25% 512 eva1aprime6x6opt -3.04087066 -6.764053 122.4% 6795% 84 lr1dr12vc10v70b-t360 8892940.708 1196600.61 86.5% 17.70% 137 fastxgemm-n3r23s5t6 27087 6087 77.5% 98.62% 901 neos-4292145-piako 112705.8 37302 66.9% 58.38% 300 allcolor58 3378 1403 58.5% 97.01% 516 allcolor10 159 67 57.9% 91.04% 92 neos-5273874-yomtsa 128.7594515 54.889718 57.4% 100% 809
- 4. Optimal Solutions: 8 Threads, 2 hrs Proved optimality on 10 models • 4 completed search (‘zero tolerance’) • 6 may have been proved to tolerance before Model Best known ODH Soln Diff Gap Time graph20-80-1rand -6 -6 0.00% 0.00% 7165 neos-4335793-snake 43 27 37.2% 0.00% 4510 neos-954925 -237.769 -237.769 0.0% 0.00% 2391 breastcancer-regularized 35.71 35.77-0.15% 0.00% 9 snp-06-004-052 1869531920 1869538126 0.0% 0.00% 474 neos-4290317-perth 3017386 3017333 0.0% 0.01% 5900 neos-3068746-nene 61910284 61910284 0.0% 0.01% 3420 gmut-76-40 -14169443 -14168137 0.0% 0.01% 141 minutedispatchstrategy 3109.9 3109.9 0.0% 0.01% 5545 gmut-76-50 -1417139 -14170547 0.0% 0.01% 1491
- 5. Useable Solutions: 8 Threads, 2 hrs Useable results on at least 7 models: • ODH gives gap < 5% • previous gap > 5% Model Best known ODH Soln Diff Gap neos-4335793-snake 43 27 37.2% 0.00% adult-regularized 7022.95 10.16 99.9% 0.01% pizza27i 773290 701882 9.2% 1.24% rmine21 -2645.00 -10500.37 297% 1.70% lr1dr04vc05v17a-t360 416992.41 255588.98 38.7% 3.04% fhnw-binschedule0 16642 16188 2.7% 3.37% bppc6-06 212 210 0.9% 4.29%
- 6. New Solutions: 8 Threads, 2 hrs Gets solutions to 5 models of 29 where no solution found before Model ODH Soln Gap kosova1 2027 100.00% neos-3740487-motru 164.4692 7.49% neos-4535459-waipa 26192754 100.00% neos-4545615-waita 8879 59.18% neos-5189128-totara 35987269 98.69%
- 7. Conclusions • Customers now want to solve larger and large models • Hard size barriers to solve (to optimality) or even to getting a solution at all • ODHeuristics can find good solutions • Useful on small(er) models too • ODH|CPLEX can provide solutions of proven optimality quality • Parallel solution methods best way of exploiting modern hardware (although limited by memory bus speeds)
- 8. Benchmarking and Evaluation • If you think that ODHeuristics and/or ODH|CPLEX might work for you: • send us your difficult matrices and we will send you the results • request an evaluation copy