Why is it slower to enter MIP start than not to enter it?
AnsweredCutting planes:
Gomory: 9
Cover: 32
Implied bound: 1
Clique: 16
MIR: 22
StrongCG: 7
Flow cover: 23
GUB cover: 59
Zero half: 11
RLT: 4
Relax-and-lift: 11
Explored 82575 nodes (3172622 simplex iterations) in 275.97 seconds (357.16 work units)
Thread count was 8 (of 8 available processors)
Solution count 9: 51 52 53 ... 741
Optimal solution found (tolerance 1.00e-04)
Best objective 5.100000000000e+01, best bound 5.100000000000e+01, gap 0.0000%
历时 289.563307 秒。
This is not input, starting from 741.
Cutting planes:
Gomory: 3
Cover: 26
Clique: 10
MIR: 17
StrongCG: 9
Flow cover: 19
GUB cover: 50
Inf proof: 2
Zero half: 2
RLT: 6
Relax-and-lift: 6
Explored 225122 nodes (7986991 simplex iterations) in 342.01 seconds (459.59 work units)
Thread count was 8 (of 8 available processors)
Solution count 4: 51 52 53 54
Optimal solution found (tolerance 1.00e-04)
Best objective 5.100000000000e+01, best bound 5.100000000000e+01, gap 0.0000%
历时 353.382064 秒。
This is using MIP Start
-
Hi,
You should run the solve with many values of Seed (I suggest at least 10 different values) with MIPstart and then without, and then compare the distributions of runtimes. A single solve of each is not enough to conclude that using MIP start is slower than not using MIP start Please see the following article for more details, and make sure to read up on performance variability and solution path (which are linked from this article):
How can I make accurate comparisons?
If you perform an accurate comparison and still find that the solve is faster without the MIP start then you should compare logs and try to understand where the difference is coming from, by asking yourself questions such as.
- Which part of the solve is taking longer?
- How does the first solution found for the solve without MIP start compare with the MIP start?
- How long does it take for the MIP start to be replaced with a better solution?- Riley
0
Please sign in to leave a comment.
Comments
1 comment