Originally posted by: Bogdan_Tanasa

Hi guys,

Is there any way to retrieve the A matrix after calling presolve. I would like to compare it with the initial matrix and see how big the differences are but I couldn't find any method for this task.

Thanks,
Bogdan.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

This can only be done in the C API or the interactive optimizer. In the C API use function CPXgetredlp to obtain the presolved problem.
In the interactive use something like this:

CPLEX> read problem.sav
CPLEX> write presolved.pre
CPLEX> read presolved.pre

After this sequence of commands the problem loaded into the interactive is the presolved problem. You can save it to a file or inspect in the interactive.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Bogdan_Tanasa

Thanks,

How it can be done in the interactive optimizer?

Bogdan,
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

But Daniel already explained it. Assuming that you have specified your problem as a file called "yourproblem.lp", you would launch the interactive CPLEX and enter:

read yourproblem.lp
write presolved.pre
read presolved.pre
write presolved.lp

This yields a file called "presolved.lp", which you can look at in your favorite text viewer/editor and compare it to "yourproblem.lp".
Tobias
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: Bogdan_Tanasa

Ah, ... I was thinking in the OPL IDE.

Thanks,
Bogdan.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Sorry, there is no way of getting an OPL model of the presolved problem.

When OPL invokes CPLEX, it constructs a flat model (i.e., unrolls the forall statements and so on). This flat CPLEX model is then subject to presolve. To get back a useful OPL model from the presolved problem would mean that you have to recover the structure and turn the flat set of constraints back to nested forall statements and the like. This is not implemented, and in some cases I doubt that this would be possible.

Of course, it would be possible to just get a flat model in OPL (without any forall or other advanced language constructs), but this would not be very useful. Looking at the LP file yields basically the same.
Tobias
#CPLEXOptimizers
#DecisionOptimization