[Top][All Lists]

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

Re: [Help-glpk] initialization of LP

From: Jinwei Gu
Subject: Re: [Help-glpk] initialization of LP
Date: Thu, 21 Feb 2008 17:31:59 -0500

Thanks a lot for your help,  xypron and Andrew. I think the problem can be solved by adding a term like |x-x_0| in my objective function where x_0 is my initial solution, as you suggested.

I looked at glp_set_row_stat and glp_set_col_stat, but I do not understand how they can be used to set initial values (say, for columns)? In these functions, I can only set the status for rows/columns to be basic (GLP_BS) or non-basic (GLP_NL).



On Thu, Feb 21, 2008 at 4:13 PM, Andrew Makhorin <address@hidden> wrote:
> I am wondering how we can do initialization in GLPK.

> In my problems, I am optimizing several thousands of small LP
> problems. I need their solutions to be similar to each other. However,
> due to various noise in these problems, some of them are not feasible
> or not easy to converge. In result, the output from GLPK for these
> problems are not close to each other at all. One solution I am
> thinking about is to use the solution from one problem as the initial
> solution for the next problem. In this way, even if the next problem
> will not converge, the result might still be close to the initial
> solution.

> Could anybody tell me how to specify the initial solution for GLPK? Or
> is it possible to do?

You can do that by specifying the initial basis which the glpk simplex
solver will start from; see api routines glp_set_row_stat and
glp_set_col_stat in the reference manual. However, this will not help
you, because there is no guarantee that the final basis obtained by the
solver will be similar to the initial one. You should formalize what
"to be similar" means in your case. Are solutions x = (0, 1, 0, .5) and
x' = (.5, 0, 1, 0) similar to each other? If so/not, why?

Best wishes,
Jinwei Gu
reply via email to

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