What kind of options can be used to reduce the crossover time while using HiGHS? #2150
-
|
Hello. I am solving an optimization problem in which the optimization using interior point solve takes about 10000 seconds as shown below. This is followed by a crossover phase which takes about 12500 seconds, even longer than the interior point solve. I wanted to know what kind of HiGHS options could be leveraged to reduce the time for crossover phase during the overall optimization, and what are the implications of tuning those parameters? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
It's not IPM crossover that takes 12500s. What happens is that after a relatively short time in crossover, the status is "imprecise", meaning that there are (residual) errors that prevent optimality from being declared. HiGHS then takes the basis after crossover, and uses it to hot start its dual simplex solver, resulting in almost a million simplex iterations in 12000s. There's no way to speed this up. All I can observe is that if you don't need a vertex solution, you can set |
Beta Was this translation helpful? Give feedback.
-
|
Many thanks for the clarification. |
Beta Was this translation helpful? Give feedback.
It's not IPM crossover that takes 12500s. What happens is that after a relatively short time in crossover, the status is "imprecise", meaning that there are (residual) errors that prevent optimality from being declared. HiGHS then takes the basis after crossover, and uses it to hot start its dual simplex solver, resulting in almost a million simplex iterations in 12000s.
There's no way to speed this up.
All I can observe is that if you don't need a vertex solution, you can set
run_crossover=off, so HiGHS will return after the IPM solve which, in this case, has obtained an optimal solution.