bknueven
commented
4 months ago Testing on the "aircond" demo:
main:
[ 0.00] Initializing mpi-sppy
[ 1.02] Initializing SPBase
[ 1.09] Initializing PHBase
[ 1.10] Starting spcomm.main()
[ 1.12] Creating solvers
[ 1.20] Entering solve loop in PHBase.Iter0
[ 1.52] Iter. Best Bound Best Incumbent Rel. Gap Abs. Gap
[ 1.52] 1 L X 1131.5603 1900.1542 67.923% 768.5938
[ 1.68] 2 L X 1135.8877 1151.6297 1.386% 15.7419
[ 1.84] 3 L 1139.2011 1151.6297 1.091% 12.4285
[ 2.01] 4 L 1141.0152 1151.6297 0.930% 10.6145
[ 2.17] 5 L 1143.1414 1151.6297 0.743% 8.4883
[ 2.35] 6 L 1144.4169 1151.6297 0.630% 7.2127
[ 2.52] 7 L X 1145.3666 1148.6334 0.285% 3.2668
[ 2.68] 8 L 1145.9572 1148.6334 0.234% 2.6762
[ 2.84] 9 L X 1146.4252 1148.0265 0.140% 1.6013
[ 3.00] 10 L 1146.5654 1148.0265 0.127% 1.4611
[ 3.17] 11 L 1146.8201 1148.0265 0.105% 1.2064
[ 3.33] 12 L X 1147.2068 1148.0099 0.070% 0.8031
[ 3.49] 13 L 1147.3424 1148.0099 0.058% 0.6675
[ 3.65] 14 X 1147.3424 1147.5520 0.018% 0.2095
[ 3.82] 15 X 1147.3424 1147.5398 0.017% 0.1973
[ 3.98] 16 X 1147.3424 1147.4736 0.011% 0.1312
[ 4.14] 17 L 1147.3560 1147.4736 0.010% 0.1177
[ 4.31] 18 L 1147.3949 1147.4736 0.007% 0.0787
[ 4.47] 19 X 1147.3949 1147.4423 0.004% 0.0474
[ 4.63] 20 X 1147.3949 1147.4235 0.002% 0.0286
[ 4.80] 21 1147.3949 1147.4235 0.002% 0.0286
[ 4.96] 22 1147.3949 1147.4235 0.002% 0.0286
[ 5.16] 23 1147.3949 1147.4235 0.002% 0.0286
[ 5.34] 24 L 1147.3972 1147.4235 0.002% 0.0263
[ 5.52] 25 X 1147.3972 1147.4123 0.001% 0.0150
[ 5.78] 26 X 1147.3972 1147.4073 0.001% 0.0101
[ 5.94] 27 1147.3972 1147.4073 0.001% 0.0101
[ 6.11] 28 L 1147.3976 1147.4073 0.001% 0.0097
[ 6.28] 29 L 1147.4028 1147.4073 0.000% 0.0045
[ 6.46] 30 1147.4028 1147.4073 0.000% 0.0045
[ 6.46] Reached user-specified limit=30 on number of PH iterations
[ 6.46] Hub algorithm PH complete, waiting for spoke finalization
[ 6.53] Spoke XhatShuffleInnerBound finalized
[ 6.60] Spoke LagrangianOuterBound finalized
[ 6.60] Statistics at termination
[ 6.60] Iter. Best Bound Best Incumbent Rel. Gap Abs. Gap
[ 6.60] 30 1147.4028 1147.4073 0.000% 0.0045
[ 6.60] Windows freed
BestInnerBound=1147.4073199141924 and BestOuterBound=1147.4028446513This PR:
[ 0.00] Initializing mpi-sppy
[ 1.05] Initializing SPBase
[ 1.13] Initializing PHBase
[ 1.14] Starting spcomm.main()
[ 1.16] Creating solvers
[ 1.25] Entering solve loop in PHBase.Iter0
[ 1.56] Iter. Best Bound Best Incumbent Rel. Gap Abs. Gap
[ 1.56] 1 L X 1131.5603 1900.1542 67.923% 768.5938
[ 1.72] 2 L X 1135.8877 1151.6297 1.386% 15.7419
[ 1.89] 3 L 1139.2011 1151.6297 1.091% 12.4285
[ 2.08] 4 L 1141.0152 1151.6297 0.930% 10.6145
[ 2.28] 5 L 1143.1414 1151.6297 0.743% 8.4883
[ 2.47] 6 L 1144.4169 1151.6297 0.630% 7.2127
[ 2.65] 7 L X 1145.3666 1148.6334 0.285% 3.2668
[ 2.86] 8 L 1145.9572 1148.6334 0.234% 2.6762
[ 3.05] 9 L X 1146.4252 1148.0265 0.140% 1.6013
[ 3.25] 10 L 1146.5654 1148.0265 0.127% 1.4611
[ 3.45] 11 L 1146.8201 1148.0265 0.105% 1.2064
[ 3.62] 12 L X 1147.2068 1148.0099 0.070% 0.8031
[ 3.80] 13 L 1147.3424 1148.0099 0.058% 0.6675
[ 3.99] 14 X 1147.3424 1147.5520 0.018% 0.2095
[ 4.16] 15 X 1147.3424 1147.4736 0.011% 0.1312
[ 4.32] 16 1147.3424 1147.4736 0.011% 0.1312
[ 4.49] 17 L 1147.3560 1147.4736 0.010% 0.1177
[ 4.66] 18 L 1147.3949 1147.4736 0.007% 0.0787
[ 4.82] 19 X 1147.3949 1147.4423 0.004% 0.0474
[ 4.99] 20 1147.3949 1147.4423 0.004% 0.0474
[ 5.18] 21 X 1147.3949 1147.4294 0.003% 0.0345
[ 5.35] 22 1147.3949 1147.4294 0.003% 0.0345
[ 5.50] 23 1147.3949 1147.4294 0.003% 0.0345
[ 5.67] 24 L 1147.3972 1147.4294 0.003% 0.0322
[ 5.83] 25 X 1147.3972 1147.4123 0.001% 0.0150
[ 6.04] 26 X 1147.3972 1147.4115 0.001% 0.0143
[ 6.20] 27 1147.3972 1147.4115 0.001% 0.0143
[ 6.37] 28 L 1147.3976 1147.4115 0.001% 0.0139
[ 6.53] 29 L 1147.4028 1147.4115 0.001% 0.0087
[ 6.70] 30 1147.4028 1147.4115 0.001% 0.0087
[ 6.70] Reached user-specified limit=30 on number of PH iterations
[ 6.71] Hub algorithm PH complete, waiting for spoke finalization
[ 6.77] Spoke XhatShuffleInnerBound finalized
[ 6.83] Spoke LagrangianOuterBound finalized
[ 6.83] Statistics at termination
[ 6.83] Iter. Best Bound Best Incumbent Rel. Gap Abs. Gap
[ 6.83] 30 1147.4028 1147.4115 0.001% 0.0087
[ 6.83] Windows freed
BestInnerBound=1147.4115020115137 and BestOuterBound=1147.4028446513Doesn't look like this change hurts multistage models.
The tests pass -- but I would like to try this on a multistage problem before merging.