Originally posted by: cgaray

Hello,

I am using the CPLEX MATLAB API and I am running an analysis that requires a change in the objective function coefficients after each iteration. Ideally, I'd like to use the previous solution from this MILP problem as a start for the next solution. I am also changing the bound on one of the rows after each iteration. The problem here is that after I make these changes to an existing problem, it seems that CPLEX no longer recognizes the existence of the objective function. After invoking cplex.solve() on the instance of Cplex containing the modified model (and after the initial problem was successfully solved) I get the following error:

Error using cplexlink123 No Cplex.Model.obj found   Error in C:\Program Files\IBM\ILOG\CPLEX_Studio_Academic123\cplex\matlab\x64_win64\@Cplex\Cplex.p>Cplex.solve (line 1474)     Error in C:\Program Files\IBM\ILOG\CPLEX_Studio_Academic123\cplex\matlab\x64_win64\@Cplex\Cplex.p>Cplex.subsref (line 3716)

The following is cplex.Model structure, showing that the objective function is indeed present:

K>> cplex.Model   ans =   sense: 
'minimization' obj: [2284x1 

double] lb: [2284x1 

double] ub: [2284x1 

double] A: [2496x2284 

double] lhs: [2496x1 

double] rhs: [2496x1 

double] name: 
'shlomi' ctype: [2284x1 

char] rowname: [2496x7 

char] colname: [2284x5 

char]

Is there a way to modify the model so that I don't have to create a new instance of Cplex each time?

Thanks,

Chris
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Eumpfenbach

Matlab is case sensitive. You are showing Cplex and cplex.

Look up the documentation for MIP starts, if you want the next problem to start with the previous solution.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: cgaray

Thanks for your response.

cplex in this case is (probably confusingly) an instance of the Cplex class. While working on this script I just used the style of the API documentation.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: cgaray

I should also add that I did re-check my scripts and that error results from calling cplex.solve(). It works fine for the initial problem but I can't generate a new solution after I modify the objective. MIP starts seem to work fine when I try them with a new model. I'd like to avoid re-creating the model every iteration, though.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Eumpfenbach

I'm just a user like you. I can't explain the problem. You might be able to skirt around it easily with something like this:

problem = Cplex()

... build your initial problem ...

problem.solve()

for i = 1:n
{

old_problem = problem;
problem = Cplex();
problem.Model.A = old_problem.Model.A;
problem.Model.rhs = old_problem.Model.rhs;
.
.
.
problem.Model.obj = {new objective function};
problem.solve()
}

It's not really efficient coding, but it's a quick fix.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: cgaray

Thanks, it seems like that may be the best way to go for now. I'll probably move this analysis into a MEX file. It looks like the modification of existing problems may be limited to the Java, C++ and C APIs: http://www.iro.umontreal.ca/~gendron/IFT6551/CPLEX/HTML/overviewcplex/modify.html. I wonder if a dev can confirm this? Is it possible that the MATLAB API could include this functionality? Maybe there are technical limitations.

Chris
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: cgaray

As an update, it seems as though I'm missing something incorrect in my code. I tried the following (based on the mipex1.m example) and it executes properly:

% Initialize the CPLEX object
cplex = Cplex('mipex1');
cplex.Model.sense = 'maximize';
      
% Use arrays to populate the model
cplex.Model.obj   = [ 1;   2;   3; 1];
cplex.Model.lb    = [ 0;   0;   0; 2];
cplex.Model.ub    = [40; inf; inf; 3];
cplex.Model.ctype = 'CCCI';
cplex.Model.A =  [-1  1  1  10;
                   1 -3  1   0;
                   0  1  0  -3.5];
cplex.Model.lhs = [-inf; -inf; 0];
cplex.Model.rhs = [  20;   30; 0];
  
% Optimize the problem
cplex.solve();
 
cplex.addRows(-Inf, [0 1 0 1], 10);
cplex.Model.obj = [0; 1; 1; 0];
 
cplex.solve()
 
% Write the solution
fprintf ('\nSolution status = %s \n', cplex.Solution.statusstring);
fprintf ('Solution value = %f \n', cplex.Solution.objval);
disp ('Values =');
disp (cplex.Solution.x);
disp ('Slacks =');
disp (cplex.Model.rhs - cplex.Solution.ax);

I've added a new row and modified the objective function coefficients. The previous solution is used as a MipStart.
#CPLEXOptimizers
#DecisionOptimization