[Symphony] Symphony with floats

Tue Apr 6 09:08:36 EDT 2010

Thanks for the follow-up.  We are still working on the conversion.  Our problem is fairly simple and we are stepping through the code now to see if we can find the point of failure.    

-----Original Message-----
From: Ted Ralphs [mailto:ted at Lehigh.EDU] 
Sent: Monday, April 05, 2010 6:03 PM
To: Schulz, Bryan
Cc: symphony at list.coin-or.org
Subject: Re: [Symphony] Symphony with floats

Just to follow up in case you're still trying, I guess it would be
extremely tricky to get rid of the use of doubles in both SYMPHONY and
its dependencies, especially the LP solver and the cut generators. I'm
not even sure it would be possible without essentially rewriting some
chunks from scratch. Ashu's suggestion to disable cut generation is a
good one, but even so, the LP solver is likely to have difficulties
with single precision. I guess if you are trying to embed the solver,
the MIPs you have must not be difficult, right? You might consider
using a lighter-weight branch-and-bound framework like the ALPS
framework provided by CHiPPS (http://projects.coin-or.org/CHiPPS) and
replacing the LP-based bounding with something simpler. I would guess
that a subgradient-based solver like the volume algorithm would be
more robust than a simplex-based solver. How accurate do you need the
solutions to be?

In any case, SymphonyTester.exe is not part of the SYMPHONY package
itself. I guess it must have been produced by some unit testing
framework or something? If you post more details like line numbers on
which the seg faults occur, we could try to help further. I'd be
curious how far you get.

Cheers,

Ted

On Tue, Mar 30, 2010 at 10:19 AM, Ashutosh Mahajan <asm4 at lehigh.edu> wrote:
> I can only speculate at this time.  Did you change the definition of DSIZE? it
> is used to allocate memory for several double arrays. Also, I would suggest
> turning off cuts just to check if the error happens because of the way cuts
> are stored.  this can be done by writing
>
> generate_cgl_cuts 0
>
> in the options file and using -f "name-of-file" in the arguments. running
> through a memory profiler like valgrind or purify can also tell where these
> errors are originating.
> --
> regards
> Ashutosh Mahajan
> http://coral.ie.lehigh.edu/~asm4
>
> On Tue, Mar 30, 2010 at 07:08:24AM -0700, Schulz, Bryan wrote:
>> We did a search and replace on doubles and changed all references to single precision (changing max and min values as well).  The changes work well for small problems, but we begin to see access violations as the problem size increases (such as for a row count of 119 and a column count of 849).  The error reported is:
>>
>> 'An unhandled exception of type 'System.AccessViolationException' occurred in SymphonyTester.exe
>> Additional information: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.'
>>
>> Any ideas where the faults could be coming?
>>
>> Thanks,
>> Bryan
>>
>>
>> -----Original Message-----
>> From: Ashutosh Mahajan [mailto:asm4 at Lehigh.EDU]
>> Sent: Friday, March 26, 2010 3:52 PM
>> To: Schulz, Bryan
>> Cc: symphony at list.coin-or.org
>> Subject: Re: [Symphony] Symphony with floats
>>
>> I don't think it will be easy to make symphony work with floats. Even if we
>> compile symphony with floats, the other packages that it depends on (CLP, CGL,
>> Osi ...) may not work with floats.
>>
>> --
>> regards
>> Ashutosh Mahajan
>> http://coral.ie.lehigh.edu/~asm4
>>
>> On Fri, Mar 26, 2010 at 12:35:58PM -0700, Schulz, Bryan wrote:
>> >    Is there a version of symphony that uses floats instead of doubles.  I am
>> >    working on an embedded application where doubles would impose unnecessary
>> >    cycles.
>> >
>> >
>> >
>> >    Thanks,
>> >
>> >    Bryan
>> >
>> >
>> >
>> >    Bryan Schulz
>> >
>> >    Algorithm Engineer
>> >
>> >    Logos Technologies, Inc.
>> >
>> >    8045 Arco Corporate Drive
>> >
>> >    Suite 230
>> >
>> >    Raleigh, NC  27617
>> >
>> >    919-926-0343  (voice)
>> >
>> >    919-272-0696  (cell)
>> >
>> >    [1]bschulz at logostech.net
>> >
>> >    [2]www.logos-technologies.com
>> >
>> >
>> >
>> >    This email message and any attachments are for the sole use of the
>> >    intended recipients and may contain proprietary and/or confidential
>> >    information which may be privileged or otherwise protected from
>> >    disclosure. Any unauthorized review, use, disclosure or distribution is
>> >    prohibited. If you are not the intended recipients, please contact the
>> >    sender by reply email and destroy the original message and any copies of
>> >    the message as well as any attachments to the original message.
>> >
>> >
>> >
>> >
>> >
>> > References
>> >
>> >    Visible links
>> >    1. mailto:bschulz at logostech.net
>> >    2. http://www.logos-technologies.com/
>>
>> > _______________________________________________
>> > Symphony mailing list
>> > Symphony at list.coin-or.org
>> > http://list.coin-or.org/mailman/listinfo/symphony
>>
>
> _______________________________________________
> Symphony mailing list
> Symphony at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/symphony
>

-- 
Dr. Ted Ralphs
Associate Professor, Lehigh University
(610) 628-1280
ted 'at' lehigh 'dot' edu
coral.ie.lehigh.edu/~ted