[BCP] Bounding in BCP

fmargot at andrew.cmu.edu fmargot at andrew.cmu.edu
Fri Feb 15 12:07:58 EST 2008


For absolute termination gap, you may use the parameter BCP_Granularity
(set by default to 1e-8). For relative termination (say with a gap of 5%), 
the simplest way is that each time a new fesible solution is found with value 
v, update the BCP upper bound to 0.95 * v instead of v.

Francois


On Thu, 14 Feb 2008, alikoc at mail.utexas.edu wrote:

> Hi all,
>
> I am trying to see how early-termination works with BCP.
>
> If I want to terminate the search tree as soon as I am within \epsilon
> neighbourhood of the incumbent, either absolute or relative, what would I do?
>
> One thing worked is returning 'BCP_DoNotBranch_Fathomed' in method
> 'select_branching_candidates()' as soon as i am within \epsilon. But, in two
> more other locations, 'BCP_lp_main_loop class' is comparing the upper and lower
> bounds, namely, after checking the true lower bound, and after generating the
> heuristic solution. In those places, I cannot instruct BCP to use my 'absolute'
> or 'relative' termination gap.
>
> Also, I tried two parameters: BCP_TerminationGap_Relative, and
> BCP_TerminationGap_Absolute, but could not succeed early termination. What are
> these parameters for?
>
> Maybe I am confused. Appreciate any help.
>
> Thanks.
>
>
>


More information about the BCP mailing list