Originally posted by: SystemAdmin

[dgravot@noos.fr said:]

Hello,


In Cplex 10, I'm trying to set up a new time limit for a MIP resolution : any time I found a new incumbent, I allow the solver to search for a new one within a user-defined time limit. It is different than the global time limit IloCplex::TiLim that can be fixed once and apply to the whole run of Cplex.

The way I managed to do so is using two callbacks : one incumbent call back that reset a timer object, and a second simplex callback. the second one calls the abort() method when the timer is over the user-defined threshold.

This probably slows down the resolution since I will call the simplex callback at each resolution of the LP. The same if I use a MIP call back that would be called at each node. Is there an other way - faster - to get the same functionality ?

Thanks


#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[achterberg said:]

You can get the same without callbacks. Just set the solution limit to 1 and the time limit to the time you want to spend after a new solution has been found. Then call CPXmipopt() in a loop, check the status indicator, and abort the loop if CPXmipopt() was not aborted due to hitting the solution limit.

#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[prubin said:]

I know the advanced start indicator is on by default, but for a MIP doesn't this restarting still incur some overhead for repeating parts of the earlier search?  The restart does not pick up with the search tree from the discovery of the previous incumbent, does it?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[achterberg said:]

[quote author=prubin link=topic=1396.msg3952#msg3952 date=1254937441]
I know the advanced start indicator is on by default, but for a MIP doesn't this restarting still incur some overhead for repeating parts of the earlier search?  The restart does not pick up with the search tree from the discovery of the previous incumbent, does it?

If you call CPXmipopt() again after it has been aborted due to some limit hit, it will continue with the existing search tree.
Therefore, you will not loose anything.

But please note that hitting a limit and resuming the search will usually have an impact on the search (in the sense that the solving process will take a different path), but this impact can be either negative, neutral, or positive; this is just random luck.

#CPLEXOptimizers
#DecisionOptimization