Originally posted by: SystemAdmin

[malak said:]

Hi all,

Just a general question about MP solvers. What does the number of iterations performed by the solver to solve a problem indicate about the problem charectaristics? I mean does a larger number of iterations say anything about the memory used or the complexity of the problem in hand ?
#DecisionOptimization
#MathematicalProgramming-General

Originally posted by: SystemAdmin

[prubin said:]

[quote author=malak link=topic=864.msg2547#msg2547 date=1232639845]
Just a general question about MP solvers. What does the number of iterations performed by the solver to solve a problem indicate about the problem charectaristics? I mean does a larger number of iterations say anything about the memory used or the complexity of the problem in hand ?


Not necessarily.  For MIPs, memory use usually increases with execution time (you're piling up live nodes in the search tree).  On the other hand, if you tell the solver to go depth-first, you'll tend to experience longer execution times than you would with a best-bound search, but you'll use a lot less memory (since only nodes along a single dive through the tree are live at any given time).

I'm not sure there's much to be inferred from execution time on a single instance of a problem, other than how lucky you are.  :-)

/Paul

#DecisionOptimization
#MathematicalProgramming-General

Originally posted by: SystemAdmin

[malak said:]

Thank you Paul for your reply-
but how about if we have two different formulations of the same instance of a problem and one is taking more time and more iterations from the solver to be solved? Can we infer anything about these two formulations?
#DecisionOptimization
#MathematicalProgramming-General

Originally posted by: SystemAdmin

[prubin said:]

You're basically do a paired difference test with a sample size of 1.  Some formulations will be tighter than others.  The formulation that's faster on your one problem instance [i]might[/i] be superior to the other formulation, but then again on a new problem instance the speed ordering might flip.  You would need a larger sample size to draw an inference with any degree of confidence.
#DecisionOptimization
#MathematicalProgramming-General