[Coin-bcpdiscuss] lots of branching compared to other solvers
hunsaker at engr.pitt.edu
Thu May 17 10:44:40 EDT 2007
We have a branch-and-price application developed with BCP. For some
testing, we turned column generation off (our code just says no
favorable columns) and tried to solve.
We found some odd behavior. BCP cannot solve in a reasonable amount of
time. There is lots of branching, but little fathoming.
For comparison, we wrote the instance to an MPS file (through BCP) and
tried it with other solvers. Both CBC and CPLEX can solve this file in
less than a minute. CPLEX can even solve it relatively quickly if cuts
and preprocessing are turned off. Both of them use far fewer nodes than
BCP; we don't even know how many nodes BCP would use since it keeps going.
Any thoughts to help us? Does this sound like an error in our BCP
interface code, or an error in BCP itself, or not an error at all?
Laci, if there's anything we can provide that would help you put us on
track we would be happy to do so. Thanks,
More information about the Coin-bcpdiscuss