[Coin-lpsolver] (no subject) (should be: possible slow solving with CLP)
John J Forrest
jjforre at us.ibm.com
Sun Aug 21 11:46:38 EDT 2005
Kish,
Thank you for finding the weakness (a free variable entered the basis not
quite as I expected). Normally I use the standalone solver "clp" and I
expect Hans Mittelmann does as well. That is a bit more sophisticated and
would probably have changed the bounds to make it go faster.
hopefully it is OK now.
Kish Shen <ks15 at icparc.ic.ac.uk>
Sent by: coin-lpsolver-bounces at list.coin-or.org
08/20/2005 05:10 PM
To
coin-lpsolver at list.coin-or.org
cc
Subject
Re: [Coin-lpsolver] (no subject) (should be: possible slow solving with
CLP)
John,
Thanks again for your reply!
Sorry - I forgot to include a subject in my original posting, as I copied
the message from a bounced mail (my original post was done in a hurry and
I
mistyped the mailing list address), and forgot to put in the subject...
>What driver are you using and what compiler?
I have tried two ways to solve the l30 problem with Clp's dual Simplex:
1)
Originally I was using the OsiCbc interface, with a call to initialSolve()
to solve a linear problem like l30 (and no integer columns).
I simply compiled all the COIN components with the default settings,
except
specifying -O and without -g for compilation.
I use gcc 2.3.4 linking with glibc 2.0 on Linux. The timings are done on
2GHz Pentium 4 Linux boxes [I compile and run on different machines --
as these 2GHz machines we have has a older gcc that has problems compiling
the COIN components, but we have a dozen of them and they are faster]
One possible difference is that I don't solve the MPS file directly, so
the order of the rows/columns may be different from when the problem is
solved from the MPS file.
2)
I then tried solving the problem using Clp's testit program, using the
`driver' driver, again compiled with the same gcc 2.3.4 configuration.
Here, with the dual solver, and solving l30 from the MPS file, Clp seems
to
get into an obvious loop, here is where it seems to start:
Clp0006I 7388 Obj -176.227 Primal inf 1.43572e+06 (419)
Clp0006I 7389 Obj -25.5449 Primal inf 1.23023e+06 (416)
Clp0006I 7390 Obj -30.2813 Primal inf 570032 (413)
Clp0006I 7391 Obj 120.83 Primal inf 659060 (409)
Clp0006I 7392 Obj 14.7871 Primal inf 543469 (415)
Clp0006I 7393 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7395 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7396 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7398 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7399 Obj 43.6816 Primal inf 1.50536e+06 (414)
Clp0006I 7400 Obj 14.7871 Primal inf 543469 (415)
Clp0006I 7401 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7403 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7404 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7406 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7407 Obj 43.6816 Primal inf 1.50536e+06 (414)
Clp0006I 7408 Obj 14.7871 Primal inf 543469 (415)
Clp0006I 7409 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7411 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7412 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7414 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7415 Obj 43.6816 Primal inf 1.50536e+06 (414)
Clp0006I 7416 Obj 14.7871 Primal inf 543469 (415)
Clp0006I 7417 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7419 Obj 23.1582 Primal inf 817259 (418)
Clp0006I 7420 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7422 Obj 43.9863 Primal inf 247895 (421)
Clp0006I 7423 Obj 43.6816 Primal inf 1.50536e+06 (414)
Clp0006I 7424 Obj 14.7871 Primal inf 543469 (415)
Clp0006I 7425 Obj 23.1582 Primal inf 817259 (418)
but there is no obvious loop for the original run (case 1, above, which I
left running for 2 days). For this case, I can generate the calls I
made to the OSI interface, which can then be compiled separately and
will hopefully reporduce the problem without using our ECLiPSe system
(which is how I have bveen using OSI). Do you want me to do this and send
the program to you?
Cheers,
Kish
_______________________________________________
Coin-lpsolver mailing list
Coin-lpsolver at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/coin-lpsolver
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/clp/attachments/20050821/1548cce8/attachment.html>
More information about the Clp
mailing list