RE: [Help-glpk] binary decision variables, --nomip

[Linder Wolfgang]

Thanks very much for your answer.
I tried with --mipgap 20 but often it still takes too long (the program has a 
lot of binary variables for diagnostic reasons, for example to calculate the 
cost if a machine has to be started, maximum up and down time etc).

Which is the option that finds the first feasible solution?
Using different weights for different levels indeed helps, but so far is not 
sufficient.

-----Message d'origine-----
De : Michael Hennebry [mailto:address@hidden
Envoyé : jeudi, 3. septembre 2009 16:56
À : Linder Wolfgang
Cc : 'address@hidden'
Objet : Re: [Help-glpk] binary decision variables, --nomip

On Thu, 3 Sep 2009, Linder Wolfgang wrote:

> I have defined binary decision variables in order to allow production only at 
> certain levels:


> plev are the allowed production levels, exactprod are the binary decision 
> variables, P is the production
>
>
>
> This works very well.
>
> However, I would like to run the program with the -nomip option because of 
> increased speed. Obviously, in this case the binary decision variables are 
> not necessarily binary any more.
>
> At most timesteps they still take the values of 0 or 1, but for example in 
> timestep 4 they don't (see below).
>
> Are there any additional conditions I could use in order to force the 
> variables exactprod to be quasi-binary while still using the -nomip option?

None that don't require a post-processing step.
IIRC there is an MIP option that says take the first fesible solution found.
That is probably closest to what you want.
There is also an MIP option that specifies
how close to optimal is good enough.

To avoid MIP altogether, you could add quadratic terms to the objective.
The terms could weight the solution away from (1/2, 1/2, ....) .

--
Michael   address@hidden
"Pessimist: The glass is half empty.
Optimist:   The glass is half full.
Engineer:   The glass is twice as big as it needs to be."