Originally posted by: Hosssein

Hi,

Could you please tell me how I can rewrite the following part of my code more efficient?

// W and S are variables

// D is a data matrix

IloNumExpr OBJECTIVE(*envvv,0);

for(i = 0; i < n; i++){

       for (j = 0; j < m; j++){

OBJECTIVE += (S[i]+D[W[i]]>j)&&(S[i]<j))* W[i];

       }            

}

Mysolve(); //It calls cp.next() in a function and solve a cp model with the objective function of OBJECTIVE.

for(i = 0; i < n; i++){

       for (j = 0; j < m; j++){

              ((S[i]+D[W[i]]>j)&&(S[i]<j)).end();

       }            

}

OBJECTIVE.end();

Actually I found out that when I delete OBJECTIVE, the part of memory related to the logical expression (S[i]+D[W[i]]>j)&&(S[i]<j)) is not deleted and I have to call "((S[i]+D[W[i]]>j)&&(S[i]<j)).end();" separately. The problem is that deleting the logical expression in this way is very time consuming.

Anybody has any idea how to write this code more efficient?

Originally posted by: Philippe_Refalo

The problem is that you recreate an expression that you end in the loop after Mysolve(). You need to store them in an array and release them after. Moreover, in Concert, constraints as expressions like (S[i]+D[W[i]]>j) need to be release by directly calling end() on them because they are not considered as expressions. Consequently, your code does not release much of the original expression

So here is a possible solution : 

1. create (S[i]+D[W[i]]>j) and (S[i]<j) for each i, j and store them in two arrays.

2. create OBJECTIVE from these arrays

3. release OBJECTIVE

4. release constraints in the 2 arrays one by one.

Regards

Philippe

Philippe

Originally posted by: Hosssein

Hi,

Thank you for your reply. Actually, I cannot use operator && to link (S[i]+D[W[i]]>j) and (S[i]<j) from the array.

IloNumExprArray   record(env);

Is there any way to handle it?

Originally posted by: Philippe_Refalo

You need to use IloConstraintArray to store the constraints. Then the operator && will be accepted.

Philippe

Originally posted by: Hosssein

Yes, it works now. Thank you.

I can see that the following line is very time consuming in my code.

Is it possible to improve the time efficiency of this line?
 

Originally posted by: Philippe_Refalo

Unfortunately, I don't see any other way to free expressions memory. An option is to avoid deleting parts of the objective that you reuse. 

Philippe

Originally posted by: rdumeur

Dear Hossein,

First, doing ((S[i]+D[W[i]]>j)&&(S[i]<j))* W[i]).end() doesn't actually delete the initial expression. It immediately deletes the newly ((S[i]+D[W[i]]>j)&&(S[i]<j))* W[i]) created expression.

If you actually need to delete the expression it would be safer to record each created ((S[i]+D[W[i]]>j)&&(S[i]<j))* W[i]) expression element into an array and then delete each array element in reverse order after solve. For that, you can't use += on the objective expression.

then after solve you can do :

Note that if you are doing very complex expressions that need to be later deleted, then it would be better to create your own  expression factory object (MyExprFactory) that would record each created expression:

I hope this helps.

Cheers,

Originally posted by: Hosssein

Thank you so much for your replies.
I face with an error on operator "&&" in line " IloNumExpr        conj(cmp1 && cmp2);".

Error: No operator && matches these operands. Could you please tell me what I can do to handle it?

Moreover, I can see you are adding some elements to record array and finally delete it without using the array. Maybe I am missing something.

Originally posted by: rdumeur

Dear Hossein

Indeed I think the type is not correct. All relational and logical expressions should be IloBoolExpr handles.

Concerning array deletion. As said in my message, *after solve*, the array is used to destroy all the created expressions by calling record[i].end().

Why do you think it is not used?

Cheers,

Originally posted by: Hosssein

OK. I got it. Thank you so much.
I tested it. Actually, since the objective is very huge, it is very time consuming to delete it at each iteration. So I have asked another related question in the forum. Could you please help me?

I am solving a CP model iteratively with same variables and constraints but different objective functions. The structure of the objective function does not change. The only things that change, from an iteration to the next, are the data arrays based on which the objective function is created in each iteration.

In my current implementation, in each iteration I delete the previous objective function and add a new objective function with new data arrays which is very time consuming.  Actually the objective is very huge and its deletion is the source of long run time.

Is it possible just to change the data arrays in each iteration inside of the objective expression to correct the objective function rather than deleting the previous objective function and adding the new objective function?

An example of such objective function is as follows: