Bug in the command line tools

52 views
Skip to first unread message

Khue Le

unread,
Feb 8, 2017, 6:47:53 PM2/8/17
to opengm
Hello,

I tried today running this:

[opengm]/build/src/interfaces/commandline/double/opengm_min_sum  -a ADSal --maxIt 1000   -m tsu-gm.h5:gm -o results/tsu-gm.h5 -v  --timeout 180 --gaplimit 0.00001 --threads 1

I set the timeout to 3 minutes, but the process just keeps running forever. I'm using OpenGM 2.3.5 (I guess this is the latest stable release?).
Could you please check?
Thank you in advance!

Khue Le

unread,
May 12, 2017, 10:05:13 PM5/12/17
to opengm, Jörg Kappes
Hi,

It seems that there is a serious bug.

[opengm]/build/src/interfaces/commandline/double/opengm_min_sum  -ADSal --maxIt 1000   -m models/triplepoint4-plain-ring-inverse.h5:gm -results/triplepoint4-plain-ring-inverse.h5 -p 100 -v  --timeout 180 --gaplimit 0.01--threads 1

Results:

Warning: Argument "memlogging" not set. Using default value: 1
Warning: Argument "precision" not set. Using default value: 0
Warning: Argument "relative" not set. Using default value: 0
Warning: Argument "lazyLPPrimalBoundComputation" not set. Using default value: 0
Warning: Argument "lazyDerivativeComputation" not set. Using default value: 0
Warning: Argument "slowComputations" not set. Using default value: 0
Warning: Argument "noNormalization" not set. Using default value: 0
running ADSal caller
begin: value inf bound -inf mem 37.676 MB
oracleCalls 37
Segmentation fault


I tried running ADSal on the benchmark and observed that it was able to solve only a (very) few instances. I just picked up some concrete instances to check and came up with the bug.
Could you please check? (Version 2.3.5.)
Thanks.

Francisco Eckhardt

unread,
May 26, 2017, 4:05:26 AM5/26/17
to opengm
Hello,

I had the same problem with the timeout (v2.3.5), but other algorithms (DDSubgradient, SRMP, TRWS and CombiLP).
time opengm_min_sum -my.h5 --timeout 1 -a DDSubgradient
[warning etc]
real    0m3.876s
user    
0m3.860s
sys    
0m0.012s


time opengm_min_sum -m my.h5 --timeout 1 -a SRMP
[warnings etc]
real    0m1.335s
user    
0m1.324s
sys    
0m0.004s

time opengm_min_sum -m my.h5 --timeout 1 -a TRWS
[warnings etc]

real    
0m1.734s
user    
0m1.720s
sys    
0m0.004s


time opengm_min_sum -m my.h5 --timeout 1 -a CombiLP --maxIt 100
[warning etc]
CPXPARAM_TimeLimit                               3600
[...]
(It seems CombiLP is not properly repassing the timeout parameter)

Any insight would be appreciated.

Best regards,
Francisco



Reply all
Reply to author
Forward
0 new messages