Re: [Help-glpk] Stop criterion

[Michael Hennebry]

On Fri, 13 Apr 2012, glpk xypron wrote:

> Hello Andrew,
>
> for assigment problems where the objective can only be integer the lower bound 
> of each node in the search tree can always be rounded to the next lower/higher 
> integer (depending on optimization direction).
>
> Probably automatic identification is difficult. It would be great if you could 
> define an option on glpsol to enable such behaviour manually.
>
> See example below.

> # kids
> set K := {1..6};
> # games
> set G := {1..6};
> # team assignment
> var t{K,G}, binary;
> # same team
> var s{K,K,G}, >=0, <=1;
> # equality measure
> var o;
>
> maximize obj: o;

Just declare o to be integer.

Note that o will have very few likely values.
The probable optimum is 2(K/2)(K/2-1)G/(K(K-1)),
where all divisions truncate,
K is the number of kids and G is the number of games.
I think that the linear relaxation will truncate to that,
but I'm not sure.

Note also, that you still have horrible symmetries.
Kids are fungible.
Games are fungible.
Teams are fungible.

I'd constrain o and minimize the deviation
from some solution selected at random.

If it helps,
any valid assignment for the first game is as good as any other.

--
Michael   address@hidden
"On Monday, I'm gonna have to tell my kindergarten class,
whom I teach not to run with scissors,
that my fiance ran me through with a broadsword."  --  Lily