[FlopCpp] Deleting a model

[Tim Hultberg]

Hi Gregory,
   I am afraid, that there is nothing you can do. There are some known memory leaks, related to the construction of the symbolic constraints. (in principle I have a fix for this but have never had the time to get it incorporated) The good news is that the leak is rather small and does only depend on the size of the symbolic model - not on the size of the generated problem. ( so with f.ex. MP_set l(4000); (if that would make sense) the increase of memory use should be the same)  

If this is a problem in practice I can think of a workaround, which could work in some cases. If the model is fixed and only the coefficients would change from loop to loop, you could generate the model only once and then use modification of the problem instance.

Tim  

-----Original Message-----
From: flopcpp-bounces at list.coin-or.org [mailto:flopcpp-bounces at list.coin-or.org] On Behalf Of gemiel at impa.br
Sent: Tuesday, September 14, 2010 6:59 PM
To: Christian Wolf
Cc: flopcpp at list.coin-or.org
Subject: Re: [FlopCpp] Deleting a model


Thanks Christian,

Unfortunatly I still observe the same phenomena : the memory used by  
the program is still increasing after each loop.
Maybe one should explicitely delete the MP_data, MP_var and  
MP_constraints after each loop-iteration ?

Gregory
_________________________________________________
#include "flopc.hpp"
using namespace flopc;
#include "OsiCbcSolverInterface.hpp"

int main() {

for (int i=1; i<100000 ; i++)
{
	MP_model loopModel(new OsiCbcSolverInterface() );
  	 loopModel.silent();
   	enum {nw,ne,se,sw,numK};

   	MP_set l(4);  // identfiers for level row and column labels  / 1*4 /;

   	MP_index i,j;

   	MP_data conc(l,l,l);

   conc(0,0,0) =  1.5;
   conc(0,0,1) =  1.5;
   conc(0,0,2) =  1.5;
   conc(0,0,3) =  0.75;

   conc(0,1,0) =  1.5;
   conc(0,1,1) =  2.0;
   conc(0,1,2) =  1.5;
   conc(0,1,3) =  0.75;

   conc(0,2,0) =  1.0;
   conc(0,2,1) =  1.0;
   conc(0,2,2) =  0.75;
   conc(0,2,3) =  0.50;

   conc(0,3,0) =  0.75;
   conc(0,3,1) =  0.75;
   conc(0,3,2) =  0.5;
   conc(0,3,3) =  0.25;

   conc(1,0,0) =  4;
   conc(1,0,1) =  4;
   conc(1,0,2) =  2;

   conc(1,1,0) =  3;
   conc(1,1,1) =  3;
   conc(1,1,2) =  1;

   conc(1,2,0) =  2;
   conc(1,2,1) =  2;
   conc(1,2,2) =  0.5;

   conc(2,0,0) =  12;
   conc(2,0,1) =  6;

   conc(2,1,0) =  5;
   conc(2,1,1) =  4;

   conc(3,0,0) =  6;

   MP_set  k(numK); //  location of four neighbouring blocks / nw,  
"ne", se, sw /
   MP_subset<3> c(l,l,l); //  neighbouring blocks related to  
extraction feasibility
   MP_subset<3> d(l,l,l); //  complete set of block identifiers

   MP_data li(k), // lead for i  / (se,sw)  = 1 /
     lj(k),     // lead for j  / ("ne",se) = 1 /
     cost(l);   // block extraction cost  / 1=3000, 2=6000, 3=8000, 4=10000 /

   li(se) = 1;
   li(sw) = 1;
   lj(ne) = 1;
   lj(se) = 1;

   cost(0) = 3000;
   cost(1) = 6000;
   cost(2) = 8000;
   cost(3) =10000;

   double value = 200000;

   forall(l*l(i)*l(j).such_that( l+i<l.size()-1 && l+j<l.size()-1 ),
	 c.insert(l,i,j)
	 );

  // c.display("c");

   forall(l*l(i)*l(j).such_that( l+i<l.size() && l+j<l.size() ),
	 d.insert(l,i,j)
	 );

   MP_variable x(l,l,l); //   extraction of blocks
   MP_constraint pr(k,c); //  precedence relationships

   pr(k,c(l,i,j)) = x(l,i+li(k),j+lj(k)) >= x(l+1,i,j);
   x.upperLimit(l,i,j) = 1;

   loopModel.maximize( sum(d(l,i,j), (conc(l,i,j)*value/100 -  
cost(l))*x(l,i,j)) );
  // loopModel.detach() ;
   cout<<"Test mine  passed."<<endl;
	}
}

Quoting Christian Wolf <kalmar at uni-paderborn.de>:

>  Hi Gregory,
>
> you could define a new MP_model for every loop.
> If you do it like this, you can not use the MP_model::getDefaultModel
> anymore to get access to the MP_model, but either the loopModel defined
> below or MP_model::getCurrentModel, which points to the last MP_model
> created.
>
>
> for (int i=1; i<100000 ; i++)
> {
>    MP_model loopModel(new OsiCbcSolverInterface() );
>
>    enum {nw,ne,se,sw,numK};
>    MP_set l(4);  // identfiers for level row and column labels  / 1*4 /;
>    MP_index i,j;
>    MP_data conc(l,l,l);
>
> ... etc ...
>
>    maximize( sum(d(l,i,j), (conc(l,i,j)*value/100 - cost(l))*x(l,i,j)) );
>    cout<<"Test mine "<<  i<<  " passed."<<endl;
>   }
>
>
> Please let me know if this helps,
> Christian
>
>
> Am 13.09.2010 23:25, schrieb gemiel at impa.br:
>> Hi,
>>
>> I have a question regarding memory management in FlopCPP.
>> Basically I am defining and solving many LPs in a loop and I would
>> like to be sure that everything is deleted after each loop iteration.
>> It does not seem to be the case since the memory used by the program
>> keeps increasing.
>>
>> I reproduced this behavior with the 'mine.cpp' example:
>> _____________________________________________________
>> for (int i=1; i<100000 ; i++)
>> {
>>    MP_model::getDefaultModel().setSolver(new OsiCbcSolverInterface);
>>
>>    enum {nw,ne,se,sw,numK};
>>    MP_set l(4);  // identfiers for level row and column labels  / 1*4 /;
>>    MP_index i,j;
>>    MP_data conc(l,l,l);
>>
>> ... etc ...
>>
>>    maximize( sum(d(l,i,j), (conc(l,i,j)*value/100 - cost(l))*x(l,i,j)) );
>>    cout<<"Test mine "<<  i<<  " passed."<<endl;
>> }
>> ________________________________________________________
>>
>> I tried to add ' MP_model::getDefaultModel().detach(); ' but it did  
>>  not help.
>> What should I do to be sure to delete all objects that are
>> instantiated in the loop ?
>>
>> Thanks.
>>
>> Regards.
>>
>> Gregory
>>
>> ----------------------------------------------------------------
>> This message was sent using IMP, the Internet Messaging Program.
>>
>>
>>
>> _______________________________________________
>> FlopCpp mailing list
>> FlopCpp at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/flopcpp



----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.



_______________________________________________
FlopCpp mailing list
FlopCpp at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/flopcpp