[Coin-discuss] Cbc-trunk goes into an infinite loop for linked mps file

Esben Mose Hansen esben at ange.dk
Fri Feb 16 06:06:19 EST 2007


Hi,

the linkedproblem makes cbc on trunk go into an infinite loop. On devel, it 
eventually runs out of memory. I have made a heuristic that takes avantage of 
the structure of the problem.,. with this, and my own driver, I can reach 
what I think is very close to the optimal solution... and then it goes into 
the same infinite loop.

The mps file can be downloaded here:
http://sirius.ange.dk/public/job-1698.mps.bz2

Any ideas to what might be the matter?

A (much shortened) log of a cbc run, though it doesn't show much. Btw, with 
the heuristic, it stops at a later node, with the same last message but Node 
159 instead. (It is a bit confusing that there are 2 node counts in play, I 
know)

esben at mosegris:~/ange-fs/externals/solver/functions/lp/coin$ cbc job-1698.mps 
loglevel 2 solv
Coin Cbc and Clp Solver version 1.01.00, build Jan 15 2007
command line - cbc job-1698.mps loglevel 2 solv
At line 1 NAME          BLANK
At line 2 ROWS
At line 2307 COLUMNS
At line 80533 RHS
At line 80945 BOUNDS
At line 93722 ENDATA
Problem BLANK has 2303 rows, 12776 columns and 137279 elements
Model was imported from ./job-1698.mps in 0.356023 seconds
logLevel was changed from 1 to 2
Optimal - objective value 3.15147e+08
Cgl0006I 15 SOS (2700 members out of 12776) with 0 overlaps - too much overlap 
or too many others
Cgl0003I 0 fixed, 520 tightened bounds, 64 strengthened rows, 0 substitutions
[...]
Cbc0015I Node 94 Obj 3.15271e+08 Unsat 20 depth 43
Cbc0015I Node 95 Obj 3.15271e+08 Unsat 18 depth 44
Cbc0015I Node 96 Obj 3.15297e+08 Unsat 16 depth 44
Cbc0015I Node 97 Obj 3.15297e+08 Unsat 13 depth 45
Cbc0015I Node 98 Obj 3.15297e+08 Unsat 17 depth 46
Cbc0015I Node 99 Obj 3.15302e+08 Unsat 9 depth 46

I have created a ticket 14 here.
https://projects.coin-or.org/Cbc/ticket/14

Thank you for the help!

-- 
kind regards, Esben Mose Hansen

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.




More information about the Coin-discuss mailing list