Originally posted by: Frops

Dear all,

I have a model constructed with OPL language.
I have some variables indexed in a set with double indices. X_ij (where i,j \in Subset)

I have exported this model to an lp file, to work with Cplex with Matlab connector.
This lp file names the variables with one index.

I would like to use a MipStart. So I need to identify the name of the variables with my original definition.

• Is there a way I can maintain the original names?
• Or how can I identify the names of the variables in the lp file with my original variables?

Thanks advance,
#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi,

the mapping is quite easy:

dvar 

int x[1..2][1..2];   subject to 
{ x[1][2]==3; 
}

gives

Minimize obj: Subject To id23: x(1)(2)  = 3 Bounds x(1)(2) Free Generals x(1)(2) End

If arrays are too big then the mapping can be less direct. To set this behaviour, you should have a look at the setting BigMapThreshold.

Regards
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Frops

Thank you AlexFleischer,

My problem had arrays too big and the mapping was not direct.

I have changed the parameter bigMapThreshold (by default 100 to 550), and generate the new lp model.
Now, I can identify the variables which maintain the name given in opl language.

But the problem remains.
I can not identify the index given to the variables, and I also need to know it to build the xml file with the solution.
Yet the name of the variables is of the type x({1,2})#3 where 3 is the index of order given to variables, in this case the third variable.

In the attached file, there is an example of some of the names of the variables, that I recognize through the i,j index. But I can not identify the order that it uses to fix the index.

Thank you,
#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi,

can you try with a much bigger bigMapThreshold ?

Regards
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Frops

Dear AlexFleischer,

I have tried with a bigger bigMapThreshold. I have noted the difference:

* With the default value (100) my variables were like this: x#484#23 without maintaining the originals i,j subindex.
* With value (550)* my variables are like this, x({1,2})#3 maintaining the subindex i,j, but with a unique index about the number of the variable in the whole model.
*The names of the variables of the lp model changes when bigMapThreshold outnumbers the dimension of the array of the variable with multiple subindexs.

The problem is that the variable is called "x({1,2})#3", so apart from knowing that the variable is the variable x({1,2}), I need to know that it has been associated with "#3" the third variable of the model. And I do not know how the assignment of the order is made.

It may seem more or less logical, but sometimes it is not. For instance, in some cases it appears a variable (I assume dummy) called id115#22, that changes all the order from expected after #22.
Thank you,
#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi,

if bigMapThreshold. is big enough, you should see all your indexes.

For example, with

dvar int x[1..2][1..2][1..300000];
 
subject to
{
 x[1][2][2]==3; 
}

with bigMapThreshold 1000000 you get

Minimize
 obj:
Subject To
 id1200027: x#0#1#1  = 3
Bounds
      x#0#1#1 Free
Generals
 x#0#1#1 
End

but with bigMapThreshold 2000000

you get

Minimize
 obj:
Subject To
 id1200027: x(1)(2)(2)  = 3
Bounds
      x(1)(2)(2) Free
Generals
 x(1)(2)(2) 
End

Regards
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Frops

Thank you again,

However, I can not manage to do what I need.

I attach you a small example to show you the problem I have. (named small)

• I wrote the mod and dat file.
• From that, I generate lp and mps model files.
• I solve this model (from both lp and mps)
• I ask for mip starts after solving the model. (solutionsmall.mst)

The final problem is that in a real case, I want to introduce an external solution through mst files (that represent mipstart solutions).
I would like to use solutions or partial solutions that I obtain from other procedures, to help CPLEX solve the problem.
I want to write a MST file with that solution, and intoduce it previously to solve execution.

MST files have the following format to give names to variables.

<variable name="x({1,2})" index="0" value="3"/>
<variable name="x({1,8})" index="1" value="3"/>
<variable name="x({5,2})" index="2" value="3"/>
<variable name="x({3,2})" index="3" value="3"/>

So, to give values to my variables in mst files, I need to identify the index="0" with variable "x({1,2})".
Although in that small case, the lp model builded from mod and dat file uses variables as expected. x({1,2})
In my real big case, I have tried to fix bigMapThreshold up to 20000000000; (that returns me a problem of Error de tiempo de ejecución de guión: cannot convert to an integer, "20000000000". Error in execution time)
For smaller values, in lp format, I get variables with this format. x({1,23})#23.
Which I understand that identify with <variable name="x({1,23})" index="23" value="0"/>
That is why I wanted to know how cplex gave the index # to the variables.
I thought I could identify the variable names in the mst file.

I am very grateful for your invaluable help,

Frops
#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi,

after the # you have the rank.
For example:

dvar int x[1..2][10..300000];
 
subject to
{
 x[2][2000]==3; 
}

gives

Minimize
 obj:
Subject To
 id600001: x#1#1990  = 3
Bounds
      x#1#1990 Free
Generals
 x#1#1990 
End

1 because 2-1=1
1990 because 2000-10=1990

so what is after the # gives you the rank

Regards
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Frops

Hi,

Thank you for clarifying that after # we found the rank.

In any case, in my model I find indexes I am not able to indentify uniquely.
I need to identify them, to generate from external procedures, a solution.
I attach two lp models as an example. Models have no fisical sense, just examples of what happens.
Note that:
-variable x, indexed in the previously build subset RR=http://1..10xhttp://1..10. 100 variables.
-variable e, indexed in http://1..10
-variable tt, indexed in http://1..10

Both lp files are generated from the same mod/dat file (also attached), with different values of bigMapThreshold.

In one case, I obtain these variables:

bigMapThreshold=100(default) ;

• x#0#0,x#1#1,x#2#2,...x#99#99 (up to 99 variables in total)
• tt(1)#100,tt(2)#102,tt(3)#104, tt(4)#106, tt(5)#108, tt(6)#110, tt(7)#112, tt(8)#114, tt(9)#116, tt(10)#118

_e(1)#101,_e(2)#103,_e(3)#105,_e(4)#107, _e(5)#109, _e(6)#111, _e(7)#113, _e(8)#115, _e(9)#117, _e(10)#119,

bigMapThreshold=1000000000; (bigger values return error)

• x({1,1})#0 , x({1,2})#1, x({1,3})#2,...,x({1,10})#9, x({2,1})#10,...,x({10,9})#98,x({10,10})#99 (up to 99 variables in total)
• tt(1)#100 and idem
• e(1)#101 and idem

I understand that after # we find an index numbering all the variables. In our case, up to 120 variables (from 0-199)
With number #0 to the first, until the last.
However, I am not able to identify the order that the export function gives to number the variable.
Thank you again,
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: Frops

Hi,

If it is not the way I try to do it, it does not matter. Another way that works is ok for me.

I would like to use externally generated solutions to help CPLEX solve the model.
I work with Cplex + Matlab, and from a model generated with OPL language and exported to MPS format.

Thank you in advance,

Frops
#DecisionOptimization
#OPLusingCPLEXOptimizer

Originally posted by: SystemAdmin

Hi All,

Does anyone know how to convert the other way (i.e.,) from LP/MPS format to OPL format???
Thanks a lot

let me know if anyone knows.

Regards,
Dinesh
#DecisionOptimization
#OPLusingCPLEXOptimizer

Hi

main
{
 cplex.importModel("hello.lp");
 cplex.solve(); 
}

Regards
#DecisionOptimization
#OPLusingCPLEXOptimizer