Originally posted by: absor
We are trying to understand column generation using OPL. for this we executed the OPL example (obj value 47) successfully. Later on we tried to solve the problem given in (example in page 7):
http://www.math.ku.dk/~tind/integercolumnWEB.pdf we have a unique observation that the knapsack subproblem was not able to generate (1 1 0) column present in the optimal solution. Instead after few iterations it generated one earlier examined patterns and with no improvement in the objective value, it fathoms.
This observation is for the case when corresponding to three demands; initial patterns were: (1 0 0) (0 1 0) & (0 0 1). However if column (1 1 0) is present among the initial patterns, then column generation is solved till optimality.
My questions are as follows:
1. Is there a dependency on the initial patterns to reach optimality (A stupid question!! I strongly feel no; as column generation being the deterministic optimization procedure, initial improved solution could make the convergence fast)
2. why the knapsack procedure is ignoring a potential patterns (our guess is the case of multiple optima/symmetry). If this stands true; what is the best way to counter this?
3. (another stupid question) as putting break points is not possible in subproblem file then how can one see subproblem formulated. We are asking for .lp kinda representation (if not possible then any workaround will be boon for us)
#DecisionOptimization#OPLusingCPLEXOptimizer