[Cmpl] Fwd: Predicting end time

[Indrajit Sen Gupta]

Hi Mike,

Thanks for the input. I realized that non linearity is what is causing all
that problem. We are trying to get access to CPLEX from our client side,
but there are some IT related issues which is causing the delay. I
installed the 64 bit CMPL on a Linux server and tried to run it on the full
problem (56 x 56 decision variables) and here is the output that I get:

CMPL model generation - running
CMPL version: 1.10.0
Authors: Thomas Schleiff, Mike Steglich
Distributed under the GPLv3
create model instance ...
cmpl: malloc.c:2842: mremap_chunk: Assertion `((size + offset) &
(_rtld_global_ro._dl_pagesize - 1)) == 0' failed.
/usr/bin/cmpl: line 115: 19491 Aborted                 (core dumped)
$dirname/bin/$appname "$cmplFile" "${cArg[1]}" "${cArg[2]}" "${cArg[3]}"
"${cArg[4]}" "${cArg[5]}" "${cArg[6]}" "${cArg[7]}" "${cArg[8]}"
"${cArg[9]}" "${cArg[10]}" "${cArg[11]}" "${cArg[12]}" "${cArg[13]}"
"${cArg[14]}" "${cArg[15]}" "${cArg[16]}" "${cArg[17]}" "${cArg[18]}"
"${cArg[19]}" "${cArg[20]}"
Is this a limitation of the architecture?

Regards,
Indrajit


On Wed, Dec 3, 2014 at 3:26 PM, Mike Steglich <mike.steglich at th-wildau.de>
wrote:

> Hi Indrajit,
>
> First of all ... I would like to suggest to use a cmplData file instead a
> Cmpl file to read your data in your Cmpl model. CmplData is the official
> data file format for CMPL. Please take a look at the attached files that I
> have created for your problem.
>
>
>
>
> The cause of the huge effort to solve your problem are the products of the
> variables in your objective function:
>
>
>
>
> *revenue: sum{i in depts, j in locs, m in depts, o in depts, n in loc1, p
> in loc2, u in level:  c[i,j] * x[i,j] +  x[m, n] * x[o, p] * adj[u, n, p] *
> rev[u, m, o] }  -> max;*
>
> This is a non-linear term. Cmpl reformulates such products of binary
> variables equivalently by a set of in-equations. Please take a look at the
> pages 46ff in the manual. The size of the model after the reformulations
> isn't very huge but the structure is quite complicated for the solver.
>
> Isn't there a different way to formulate your model and to avoid products
> of variables. It seems that it is a multistage transportation model, for
> that it could be possible to find alternative formulations.
>
> If you can't reformulate it then I would like to propose to use Cplex or
> Gurobi. You can obtain a free version via the IBM academic initiative or
> via the Gurobi web page.
>
> https://www.ibm.com/developerworks/community/blogs/jfp/entry/cplex_studio_in_ibm_academic_initiative?lang=en
> http://www.gurobi.com/products/licensing-and-pricing/academic-licensing
>
> For both solvers Cmpl's solutionPool feature can be used that can be
> combined with a mip gap to decrease the computation times.
> You can use the following Cmpl header entries for Cplex:
> *%arg -solver cplex*
> *%opt cplex mip/tolerances/mipgap 0.10*
> *%display solutionPool*
>
>
> Cheers,
>
> Mike
>
>
>
>
>
>
> Am 03.12.2014 um 03:00 schrieb Indrajit Sen Gupta <indrajitsg at gmail.com>:
>
> The optimizer finally completed after about 5.3 hours. Attached are the
> files. It is a store space optimization problem which I am trying to solve.
> As a proof of concept, this data has been reduced to 1/5 th of the actual
> problem.
>
> Is there way to reduce the time it takes or generate intermediate outputs
> so that we don't have to wait till it completes?
>
> Regards,
> Indrajit
>
>
> On Wed, Dec 3, 2014 at 2:34 AM, Mike Steglich <mike.steglich at th-wildau.de>
> wrote:
>
>> Hi Indrajit,
>>
>> Can you send the model for a better understanding?
>>
>> Thanks,
>>
>> Mike
>>
>>
>>
>> Am 02.12.2014 um 17:38 schrieb Indrajit Sen Gupta <indrajitsg at gmail.com>:
>>
>> Hi Mike,
>>
>> I am running CMPL on a small version of very big problem I need to solve.
>> The current problem has 100 binary variables and about 20 constraints. The
>> objective function is a bit non linear as it has a term x[i,j] * x[k, m].
>> The program has been running for about 3 hours now and the output is
>> showing:
>>
>> After 20300 nodes, 5239 on tree, -51762900 best solution, best possible
>> -62975151 (10407.49 seconds)
>>
>> Any idea how much time it will take to complete? I understand, with
>> binary variables it is difficult to predict, but supposing I were to stop
>> in mid way, is there any way to get the current best solution?
>>
>> Regrads,
>> Indrajit
>>
>> _______________________________________________
>> Cmpl mailing list
>> Cmpl at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/cmpl
>>
>>
> <prob008.cmpl><prob008data.cmpl>
>
>
>
>
> _______________________________________________
> Cmpl mailing list
> Cmpl at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/cmpl
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cmpl/attachments/20141203/11378a73/attachment.html>