<html style="direction: ltr;">
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
<style type="text/css">body p { margin-bottom: 0cm; margin-top: 0pt; } </style>
</head>
<body style="direction: ltr;"
bidimailui-detected-decoding-type="latin-charset" bgcolor="#FFFFFF"
text="#000000">
<div class="moz-cite-prefix">Yes, different factorization routines
*should* give the same answer (barring numerical instability);
it's just a question of speed. <br>
<br>
Taucs is pretty dated by now, but you should get reasonable
performance from Mumps or CHOLMOD. Unfortunately I can't be
helpful with compilation issues on win32. You might find it easier
to use IPOPT, which also has interfaces to state-of-the-art HSL
routines (free for academics) and PARDISO. <br>
<br>
Miles<br>
<br>
On 06/22/2012 04:20 AM, Petru Pau wrote:<br>
</div>
<blockquote cite="mid:4FE438C0.3060703@risc.uni-linz.ac.at"
type="cite">With the native factorization, CLP solves with barrier
a rather big LP (12397 rows, 767497 columns, 6704873 elements) in
more than 59 000 seconds.
<br>
<br>
CPLEX and Gurobi solve the same LP in 130 and 100 seconds,
respectively.
<br>
<br>
Dense factorization does not seem to bring any improvement.
<br>
<br>
I tried all alternative factorizations mentioned in the FAQ. I was
able to compile and link AMD and Taucs (on Windows, with Microsoft
Visual Studio compiler). For Mumps I need a Fortran compiler, for
CHOLMOD I could not find makefiles for win32, and WSMP is not
(and will not be) freely available for win32.
<br>
<br>
Anyway.
<br>
<br>
No matter what factorization I use, the iterations reported by CLP
give the same numerical values, in lines like:
<br>
<br>
16 Primal 50944522 Dual -424542454 Complementarity 5523133 - 0
fixed, rank 12397
<br>
<br>
I know that the chosen factorizations is employed, I inserted
messages at the beginning and end of order(), factorize() and
solve() functions in ClpCholeskyTaucs.cpp and ClpCholeskyUfl.hpp,
and these messages are displayed.
<br>
<br>
I have no knowledge about barrier method, but I am a bit
perplexed: Is it supposed to be this way? Is the factorization
needed only for, I don't know, computing faster an iteration? (In
any case, 18 iterations with Taucs took longer than 39 iterations
with native factorization.) It seems that no matter what
factorization I use, CLP follows the same trajectory in its search
for a solution.
<br>
<br>
If the answer to the question above is rather immediate after a
quick search in documentation, I apologize. Yet there is still a
question to ask: Is there any chance to solve with CLP, in
reasonable time, similarly big LPs?
<br>
<br>
Regards,
<br>
Petru
<br>
_______________________________________________
<br>
Clp mailing list
<br>
<a class="moz-txt-link-abbreviated" href="mailto:Clp@list.coin-or.org">Clp@list.coin-or.org</a>
<br>
<a class="moz-txt-link-freetext" href="http://list.coin-or.org/mailman/listinfo/clp">http://list.coin-or.org/mailman/listinfo/clp</a>
<br>
</blockquote>
<br>
<br>
</body>
</html>