Originally posted by: Ibrahim_Capar

Hello all,

I am using lazy constraints to add complicating constraints to my problem. Before add a new constraint, I check whether the current solution is violating any lazy constraint.  Let say that I have only one set of decision variable, "X", which is binary.  I use:

getValues(XSol, X); to copy current solution (one dimensional array) . Naturally XSol might be epsilon away from 0 or 1, i.e. 0.00001 is 0 and 0.99999 is 1;. However I realized that XSol contains values such as:

0.2
0.333333
0.5
0.6
0.666667
0.8

From my experimentation, this occurs on CPLEX 12.6.2 (64 bit), and 12.6.3 (64 bit). When I run the same code with 12.5.1 (64 bit), everything works fine.

I am using C++ on Microsoft Visual Studio Professional 2012 Ver 11.0.6 Update 5 under Windows 7 64 bit. My code structure is:

ILOLAZYCONSTRAINTCALLBACK1(UserCutName, IloIntVarArray, X){

...

getValues(XSol, X);

...

}

int main(){

....

IloEnv env;

IloIntVarArray X(env);

for (int i=0; i<Node; i++)
  X.add(IloIntVar(env,0,1));   

.....

IloCplex cplex(env);

cplex.use(UserCutName(env, X));

.....

}

 

I appreciate it any help to fix this problem. Thank you in advance.

Ibrahim

#CPLEXOptimizers
#DecisionOptimization

This indeed looks odd. The lazy constraint callback should only be invoked with solutions that are integer feasible.

I take it, you only observe this in LazyConstraintCallbacks, not in UserCutCallbacks? Are the values already fractional if you call getValues() first thing in your callback?

#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Ibrahim_Capar

I did not try it with UserCutCallbacks.  I extracted values of X with following code.

 

ILOLAZYCONSTRAINTCALLBACK1(UserCutName, IloIntVarArray, X){
ofstream output("output.txt", std::ofstream::app);
IloEnv env = getEnv();
IloNumArray XSol(env, Xsize);
getValues(XSol, X);
    for(int i=0;i<Node;i++){
            if ( XSol[i] > 0.01 &&  XSol[i] < 0.95) output << XSol[i] << endl;
            if ( XSol[i]< -0.01 ||  XSol[i] > 1.01) output << XSol[i] << endl;
        }

.....

}

#CPLEXOptimizers
#DecisionOptimization

Can you double-check that XSol[i].getType() is indeed binary (maybe print this as well)? If that is the case then I have no clue what is going wrong. Do you have a small sample code that reproduces the issue and that I could run here?

#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Ibrahim_Capar

I cannot use XSol[i].getType(). It says "Error: expression must have class type". However I was able to use it for X[i].getType(); I have following code and output.

for(int i=0;i<Node;i++){
    if ( XSol[i] > 0.01 &&  XSol[i] < 0.95) output << XSol[i] << ":" << X[i].getType() << ":" << getValue(X[i])  << endl;
    if ( XSol[i]< -0.01 ||  XSol[i] > 1.01) output << XSol[i] << ":" << X[i].getType() << ":" << getValue(X[i])  << endl;
}

0.8:3:0.8
0.2:3:0.2
0.2:3:0.2
0.2:3:0.2
0.6:3:0.6
0.8:3:0.8
0.2:3:0.2
0.6:3:0.6
0.6:3:0.6
0.2:3:0.2
0.8:3:0.8

Getting a sample and related code, unfortunately, will take some time.

#CPLEXOptimizers
#DecisionOptimization

Sorry, my bad. I meant to say X[i].getType().

One more thing to double-check would be to make sure that Node<=XSol.getSize(). I guess this implied by your code but could you still check in the loop in which you print the values that i<XSol.getSize()?

I have run a code similar to yours on quite a number of models here and so far I did not see any fractional values in a lazy constraint callback. So I am afraid there is not much I can do without looking at your actual code :-(

#CPLEXOptimizers
#DecisionOptimization