Re: [Help-glpk] order restrictions

[glpk xypron]

Hello Robin,

> I have a MIP, very similar to the sudoku.mod example in the examples/
> directory.
MIP optimzers are not fast solvers for constraint programming problems.

To choose the variable glpsol uses the Tomlin-Driebeek heuristic by default, cf.

http://opim.wharton.upenn.edu/~guignard/916_2011/papers_to_read/Driebeck_1966_MSc.pdf

and

*  Tomlin J.A. Branch and bound methods for integer and non-convex
*  programming, in J.Abadie (ed.), Integer and Nonlinear Programming,
*  North-Holland, Amsterdam, pp. 437-50 (1970).

You can influence the branching variable if you use GLPK library and provide a 
callback function. See
gpk-4.47/doc/glpk.pdf

Best regards

Xypron

-------- Original-Nachricht --------
> Datum: Thu, 1 Nov 2012 11:20:24 +1300
> Von: robin hankin <address@hidden>
> An: address@hidden
> Betreff: [Help-glpk] order restrictions

> hello.
> 
> I have a MIP, very similar to the sudoku.mod example in the examples/
> directory.
> 
> It's a 3D binary array.
> 
> Let's say I have three restrictions, A, B, and C.
> 
> A is satisfied 70% of the time and it takes a long long time to evaluate.
> B is very similar to A but is satisfied 80% of the time.
> C is satisfied 0.0001% of the time and is very very quick.
> 
> A human would ignore A and B, and concentrate on C.  When he found a
> proposed solution
> that satisfies C, he would then check A and B (and if either of these is
> not satisfied he would
> go back to trawling through C).'
> 
> 
> How do I tell glpsol this information about the best order to evaluate
> A/B/C?  Or does it "know"?
> 
> 
> 
> 
> 
> 
> -- 
> Robin Hankin
> Uncertainty Analyst
> address@hidden