bug-glpk
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-glpk] glp_warm_up


From: Andrew Makhorin
Subject: Re: [Bug-glpk] glp_warm_up
Date: Thu, 07 Mar 2013 08:30:01 +0400

I fixed the bug. Please see a patched version of glpapi12.c attached.
You need to replace that file and rebuild the package as usual. (The
patch is for v.4.48.)

> My application involves optimizing a problem, making a small change and 
> then optimizing again. I use warm-up to see if the modified problem is 
> still optimal. If that is the case then I can bypass a lot of steps. So 
> the test is critical for my application. 

Then I'd suggest you to call glp_warm_up and lpx_check_kkt; the latter
provides a more robust primal/dual feasibility test.

> I am currently using glpk-4.36. 
> I found the problem when I tried some of my test cases using 4.48. Most 
> of the ones that involved the use of warm-up failed to produce the 
> expected answers.(The algorithm was skipping steps that needed to be 
> performed.) Since the problems involved thousands of variables and 
> constraints, I worked up the simple example to send instead. (I verified 
> that I had found the problem by hacking the test in warm-up to look at 
> the variable status as well as type.)
> 
> In my application some variables are double bounded. Occasionally the 
> two bounds are the same and I set the type in glpk to fixed. However, 
> you need to write your tests in warm-up so that they work correctly for 
> problem instances that say the variable has a double bound but it is in 
> fact fixed.
> 
> I have a technical question about scaling for the dual feasibility test. 
> I am careful to scale the rows and columns in the constraint matrix so 
> that the largest (absolute value of an) entry in each row and column has 
> order of magnitude 1. Should the objective row be scaled in the same way?

It is desirable, but not necessary, because the simplex solver performs
such scaling internally.

> 
> I very much appreciate the effort that you have put into glpk and do 
> find that it is a powerful tool for the problems that I work on.

Thank you for your interest in glpk.


Andrew Makhorin

Attachment: glpapi12.c.gz
Description: GNU Zip compressed data


reply via email to

[Prev in Thread] Current Thread [Next in Thread]