Originally posted by: YongjiaSong

Hi all,

I have a problem in iteratively solving the LP. In each iteration, I have to modify the objective coefficient as well as variable bounds.

How can I implement this using C++ in CPLEX?

I have done an easy version of it by only changing the objective coefficient. What I did is:

IloModel singleLP(env);
IloCplex singleLP_Cplex(singleLP);
IloNumVarArray x(env, knap.nbItems, 0, 1);
IloObjective objective(env, IloObjective::Maximize);
singleLP.add(objective);
IloExpr str(env);
for (int j = 0; j < knap.nbItems; ++j)
str += x[j]*knap.weight[i][j];
singleLP.add(str <= knap.capacitykk);
singleLP_Cplex.setParam(IloCplex::SimDisplay, 0);
singleLP_Cplex.setParam(IloCplex::PreInd, 0);
While (iter < MaxIter)
{
IloObjective oldobjective = singleLP_Cplex.getObjective();
singleLP.remove(oldobjective);
IloExpr newobj(env);
for (int j = 0; j < knap.nbItems; ++j)
newobj += x[j]*knap.weightiihttp://k*knap.nbItems+j;
singleLP.add(IloMaximize(env, newobj));
singleLP_Cplex.setParam(IloCplex::SimDisplay, 0);
try{
singleLP_Cplex.solve();
}
iter++;
}

Which removes the old objective and a new one. I tried to use objective.setExpr(), but it doesn't work.
Thanks!
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: GuangFeng

Please take a look at the summary of problem modification APIs here

If you just need to change linear coefficients in the objective, please consider using IloObjective::setLinearCoef() or IloObjective::setLinearCoefs().
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

How did you use IloObjective::setExpr()? I just tried

cplex.getObjective().setExpr(...);

and that worked without a problem for me.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: YongjiaSong

Hi,

I think my problem is that I initialize the IloObjective with an objective without an expression, I just set the sense. Later in each iteration, when I set the expression, it doesn't work.

So now I fix that problem, but I find that iterative LP runs very slow, the LP's are quite small, like 50 variables and 5 constraints. How do I warm up the LP for the next iteration by using the information we've got from previous iteration? Does CPLEX do it automatically or I have to store the optimal basis or something like that?

Thanks!
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

> YongjiaSong wrote:
> Hi,
>
> I think my problem is that I initialize the IloObjective with an objective without an expression, I just set the sense. Later in each iteration, when I set the expression, it doesn't work.
>
Note that a statement like

IloObjective objective(env, IloObjective::Maximize);

will invoke this constructor:

IloObjective(

const IloEnv env, IloNum constant=0.0, IloObjective::Sense sense=Minimize, 

const 

char * name=0)

i.e., it will create an objective function with constant -1 (the numerical value of IloObjective::Maximize) and sense 'minimize' -- not what you intended to do. In order to create an empty objective function with sense 'maximize' use

IloObjective objective(env, 0, IloObjective::Maximize);

> So now I fix that problem, but I find that iterative LP runs very slow, the LP's are quite small, like 50 variables and 5 constraints. How do I warm up the LP for the next iteration by using the information we've got from previous iteration? Does CPLEX do it automatically or I have to store the optimal basis or something like that?
>
CPLEX will automatically try to use all information that is available from the previous solve. What kind of changes do you apply to the model in the iterations?
#CPLEXOptimizers
#DecisionOptimization