RE: [Help-glpk] optimality tolerance

[Ciyou Zhu]

1.  All these parameters are explained in detail in GLPK documents.

2.  The solution time could range from few minutes (if not seconds) to few
days (if not longer) based on your parameter setting and the speed of your
machine. 

3.  I have tried on even bigger problems, i.e. with around 1,000,000
variables/3000 constraints.  Version 3.2.3 begins to suffer from numerical
instability in the LP phase.  Version 4.4 seems suffer less.  But I haven't
had time to explore 4.4 further yet. 

Ciyou Zhu
www.ece.nwu.edu/~ciyou


-----Original Message-----
From: address@hidden
[mailto:address@hidden On Behalf Of Dedalus
Sent: Friday, February 20, 2004 11:05 AM
To: address@hidden
Subject: RE: [Help-glpk] optimality tolerance

Ciyou,

I am very interested, could you please expand on this ?

LPX_K_TOLOBJ should be the object tolerance parameter, but what's the use of
LPX_K_TOLPIV and LPX_K_TMLIM ?

>I have been using GLPK to solve IP problems upto 300,000 variables/1000 
>constraints derived from crew pairing optimization.

what is the answer time of GLPK for your big model, after your changes ?

Thank you

Duilio


>I did it for GLPK-3.2.3.  (To input LPX_K_TOLOBJ, LPX_K_TOLPIV and 
>LPX_K_TMLIM as environmental variables.)  You need to edit glpsol.c
for that
>purpose.  You need also to edit glplpx3.c if you want a TOLOBJ >
0.001.  See
>attached files.
>
>I have been using GLPK to solve IP problems upto 300,000 variables/1000 
>constraints derived from crew pairing optimization.
>
>Ciyou Zhu
>
>-----Original Message-----
>From: address@hidden
>[mailto:address@hidden On Behalf Of
Michael
>Hennebry
>Sent: Thursday, February 19, 2004 9:58 PM
>To: Brady Hunsaker
>Cc: address@hidden
>Subject: Re: [Help-glpk] optimality tolerance
>
>On 19 Feb 2004, Brady Hunsaker wrote:
>
>> On Mon, 2004-02-09 at 10:49, Duilio Foschi wrote:
>> > is there a way to set the optimality tolerance of a fairly large
MIP 
>> > problem in GLPK ?
>> >
>>
>> I do not believe there is currently a way to do this.  This should
be 
>> easier with the Integer Optimization Suite that is being developed, 
>> but for the time being, large MIPs are probably not appropriate for
GLPK.
>
>There is a parameter for it, but to do what you want you will probably
have
>to edit the glpk source.
>It appears to be intended to deal with the vagaries of approximate 
>arithmetic.
>Accordingly a value larger than .001 is flagged as an error.
>
>-- 
>Mike   address@hidden
>"Two roads divurged in a woods, and I took the road less travelled
by...
>and they CANCELLED MY FRIKKIN' SHOW. I totally shoulda took that road
>with all those people on it. Damn."                           --  Joss
>Whedon
>
>
>
>_______________________________________________
>Help-glpk mailing list
>address@hidden
>http://mail.gnu.org/mailman/listinfo/help-glpk
>
>
Attachment: glpsol.cAttachment: glplpx3.c>
>_______________________________________________
>Help-glpk mailing list
>address@hidden
>http://mail.gnu.org/mailman/listinfo/help-glpk
>
>
>
>
>

       

-----------------------------------------------------
http://eo.yifan.net
Free POP3/Web Email, File Manager, Calendar and Address Book


_______________________________________________
Help-glpk mailing list
address@hidden
http://mail.gnu.org/mailman/listinfo/help-glpk