Originally posted by: SystemAdmin

[Sylvain said:]

I would like to solve successively many similar LPs (MIPs with fixed integer variables).
I see two ways of doing it.

First approach:
  1. Write a generic model
  2. Add specific constraints to the model
  3. Extract and Solve
  4. Remove specific constraints from the model
  5. Return to stop 2.

Second approach:
  1. Write a generic model
  2. Extract the model
  3. Add specific constraints as cuts using IloCplex::addCuts()
  4. Solve
  5. Clear cuts using IloCplex::clearCuts()
  6. Return to step 3.

The second approach seems to be faster as the model is extracted only once (although adding cuts will modify the search algorithm).
I have tried this approach but it returns different solution then the first one.

Does anyone have any comment or suggestion to help me ?
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[prubin said:]

When you say "different solution" do you mean the same objective value (to within rounding error) but a different corner, or a different objective value (with the difference too large to account for by tolerances)?

I'm not positive, but I suspect that the presolver ignores lazy constraints (which is what addCut adds).  That could account for differences in the sequence of pivots and maybe, if the problem is stiff, even differences in the objective value.

/Paul
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

[Sylvain said:]

The difference is in term of the objective function and it's quite noticeable (about 10%)
#CPLEXOptimizers
#DecisionOptimization