RE: [Help-glpk] Using branch and cut starting with relaxed solution

[Joey Rios]

Hi Ali,

> Maybe i am missing some important info.
>
> Why don't you just set the status of the rows and columns
> (glp_set_col_stat and glp_set_row_stat) then call lpx_warm_up?
>

This was one of my initial questions.  How to decide which rows/cols should be basic?  I think that the relaxed optimal solution I am getting isn't at a corner point (kind of like how a barrier method would provide an optimal solution?).  As such, I need to figure a way to use my initial solution to come up with a basis (I think).  I'd even be happy if that initial basis wasn't optimal.  If I had to let glp_simplex run for a little while (a few mins?) from whatever basis I provided to re-find an optimum, I'd be OK with that.  It seems, though, that deciding which rows/cols should be basic given my relaxed solution is quite non-trivial. 

I don't know if I mentioned it before or not, but all decision variables are binary (it's a pure 0-1 program).

> If you are problem is so big that solving the initial relaxation lasts
> for several days, are you sure you can solve the corresponding MILP
> problem in acceptable time?

Luckily, the formulation of the problem is such that the relaxed solution found through simplex is quite often an integer solution already.  And when it isn't, there (usually) isn't much branching necessary.

Joey

---

Send e-mail anywhere. No map, no compass. Get your Hotmail® account now.