headers
Andrew Makhorin | 14 Nov 2002 05:07
Picon

GLPK 3.2.3 release information

GLPK 3.2.3 -- Release Information
*********************************

Release date: Nov 11, 2002

GLPK (GNU Linear Programming Kit) is intended for solving large-scale
linear programming (LP), mixed integer linear programming (MIP), and
other related problems. It is a set of routines written in ANSI C and
organized in the form of a callable library.

A new implementation of the api routine lpx_integer which now is based
on the b&b driver (which is based on the implicit enumeration suit) was
included in the package. This new implementation has exactly the same
functionality as the old version, so all changes are transparent to the
api user.

Four new api routines were included in the package:
lpx_check_kkt  check Karush-Kuhn-Tucker optmality conditions;
lpx_read_bas   read predefined basis in MPS format;
lpx_write_bas  write current basis in MPS format;
lpx_write_lpt  write problem data in CPLEX LP format.

Three new command-line options were introduced to the solver glpsol:
--plain (use plain names), --orig (use original names), and --wrlpt
(write problem data in CPLEX LP format).

Also other minor improvements were made (see the file 'ChangeLog').

Please note that the current version of GLPK is *tentative*.
(Continue reading)

Permalink | Reply | 0 comments |
headers
odwl | 15 Nov 2002 16:46
Picon

Re: GLPK - GNU Project - Free Software Foundation (FSF)

Hi,

It would be nice if we could specify a limit of computing time 
(particulary for MIP problems)  and obtain the best founded solution 
after this lapse of time.

Thanks for your job.

Olivier
Permalink | Reply | 1 comment |
headers
Andrew Makhorin | 15 Nov 2002 22:45
Picon

Re: [Help-glpk]

>It would be nice if we could specify a limit of computing time 
>(particulary for MIP problems)  and obtain the best founded solution 
>after this lapse of time.

You can limit the computing time by setting the control parameter
LPX_K_TMLIM via the api routine lpx_set_real_parm. (For details please
see the reference manual.)

If you mean the solver glpsol, there is no such possibility. However,
you can limit the time, for example, to ten minutes (600 seconds) by
inserting the statement:

      lpx_set_real_parm(lp, LPX_K_TMLIM, 600.0);

before the comment line 420 (file sample/glpsol.c, glpk 3.2.3):

      /* solve the problem */
Permalink | Reply | 0 comments |
headers
Sebastien.deMentendeHorne | 19 Nov 2002 13:52

avoiding destruction of matrix with lpx_load_mat3

Hi,

the function lpx_load_mat3 removes all previously defined coefficients of
the constraint matrix. is there a way to avoid this (like CPXchgcoeflist in
CPLEX) ?

tx
Permalink | Reply | 1 comment |
headers
Andrew Makhorin | 19 Nov 2002 15:48
Picon

Re: avoiding destruction of matrix with lpx_load_mat3

>the function lpx_load_mat3 removes all previously defined coefficients of
>the constraint matrix. is there a way to avoid this (like CPXchgcoeflist in
>CPLEX) ?

To change the constraint matrix you can use the routines lpx_set_mat_row
(to replace a row) or lpx_set_mat_col (to replace a column).

/* Sebastien, my e-mails I sent you a while ago didn't reach you because
of a delivering problem on your mail server. - A.M. */
Permalink | Reply | 0 comments |
headers
Andrea Curtoni | 19 Nov 2002 19:13
Picon
Favicon

Using a heuristic to find a first integer solution quickly

I found on the net this message:

On Tue, 15 Oct 2002, Dai, Jianrong wrote:

>> Hi, everyone,
>>
>> I need to solve a MILP problem with hundreds of binary variables. Because
>> GLPK runs too slow (already 4 days, still running), I'm thinking about using
>> a optimal solution of an heuristic method to speed up the optimization
>> process. I'm wondering whether it is possible to do so. If the answer is
>> yes, please show me how to do it. Thanks.
>>
>> Jianrong
>>
>It should be possible. One (not very good way) might be to use the
>"objective value" from the heuristic method and modify the source of the
>branch and bound solver so that it is possible to specify a "cut" cost
>(eg. set the variable bb->best (?) to the objective value from the
>heuristic method. This should make the bb solver to skip all branches
>that is more "expansive" than the heuristic solution.
>
>/Niklas

I have a similar problem, GLPK takes too much time to find the first integer feasible solution, but i have a
heuristic able to find a sub-optimal integer feasible solution in a short time.
How can I use it?
Isn't there a simpler method than changing the source of the branch and bound solver?

Thanks in advance,
Andrea
(Continue reading)

Permalink | Reply | 627 comments |
headers
Andrew Makhorin | 19 Nov 2002 20:28
Picon

Re: Using a heuristic to find a first integer solution quickly

>I have a similar problem, GLPK takes too much time to find the first
>integer feasible solution, but i have a heuristic able to find a
>sub-optimal integer feasible solution in a short time.
>How can I use it?
>Isn't there a simpler method than changing the source of the branch and
>bound solver?

A simpler method must appear in glpk 3.3 :+)

In glpk 3.2.3 you can insert two statements between lines 634 and 635 in
the file glplpx6c.c as follows:

633:  /* reset the status of the mip solution */
634:  mip->i_stat = LPX_I_UNDEF;
      tree->found = 1;
      tree->best[0] = <objective value from a heuristic>;
635:  /* solve the problem */
636:  ret = mip_driver(tree);
Permalink | Reply | 0 comments |
headers
Andrew Makhorin | 19 Nov 2002 20:39
Picon

Re: Using a heuristic to find a first integer solution quickly

>Isn't there a simpler method than changing the source of the branch and
>bound solver?

I agree that changing the source code is not a good practice. However,
you can copy, say, the file glplpx6c.c to your working directory,
rename it, and compile it with your program in order not to change and
not to re-install the glpk package itself.
Permalink | Reply | 627 comments |
headers
Gilberto de Miranda Junior | 27 Nov 2002 15:26
Picon

Multidimesional Arrays in GLPK/L


Hi There,

I am a teacher working with operations research and i am
using GLPK as a base for the homeworks  of my students.

Since many of them are not familiar with C programming, i choose
to teach them about GLPK/L.

In the last homeworks, one of my students asks me a question that i
was not able to solve:

How to work with cost matrices that have more than two dimwnsions?

This question arises in many problems involving multicommodity network
flows.

If possible, can you make this point clear to me?

Thanks for your work!

Greetings,

Gilberto de Miranda
Permalink | Reply | 3 comments |
headers
dm2118 | 27 Nov 2002 17:12
Favicon

linking with msvc

Hallo,

I would like to use GLPK with Micorsoft Visual Studio.

I followed the instructions of Andrew Makhorin in the message of 
the Thu, 18 Jul 2002 and I created the dll's with the 
"makedll.bat" which he suggested.

Still am not able to link.

It would be very nice if somebody could help me, and tell me the 
options that I have to chose in order to link succesfully (actually 
I am also pretty novice to Visual studio too.)

Thank you very much.

Danilo Mercurio
Permalink | Reply | 0 comments |

Gmane