[Coin-symphony] Fix the variables
Yossiri Adulyasak
a.yossiri at gmail.com
Sun May 28 00:53:11 EDT 2006
Hi Ted,
Thanks for your reply. You understood correctly. In essence, we try to use
column generation to solve an IP without using Branch-and-Price. That is,
we generate columns outside of B&B tree.
Here's an outline of our column generation algorithm:
1. Formulate a Restricted Master Problem (RMP) with subset of columns from
the original problem.
2. Solve the above RMP in B&B to IP optimal.
3. Use the optimal IP solution, to price and generate columns.
4. Solve the RMP + additional columns generated in B&B to optimality.
5. Repeat 3 and 4, until no more columns are generated.
Now, in Step 3, we need dual information to price the columns. And I guess
here's where I confused you. I used to work with CPLEX; in CPLEX it allows
users to obtain dual info of the optimal IP solution (or more correctly, of
the LP relaxation solution at the optimal B&B node) by changing the problem
type from IP to LP-fixed, in which, all variables are fixed at the IP
solution. I think what CPLEX does (to fix those variables) is to solve LP
relaxation with the conditions at that B&B node to reproduce the integer
solution and hence dual info. That is, it does not literally fix the
variables. (So, I confused you when I said we tried to fix the
variables.)
So, the question is, can we do the same thing with SYMPHONY/CLP?
Many thanks for your help. I appreciate you time and help.
Cheers,
> --
> -Yossiri Adulyasak- Graduate Student
> Transportation engineering Program
> Department of Civil Engineering
> Chulalongkorn University
> Thailand
>
a.yossiri -at- gmail.com
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/symphony/attachments/20060528/dde4d586/attachment.html>
More information about the Symphony
mailing list