Originally posted by: SystemAdmin

[alan_a said:]

I have a cplex allocation problem.
One of my constraints is that I'm allocating tuples of y to tuples of x.
Each y is unique, there can be up to n y values in each x.

There is a concept of grouping of y such that (and forgive the javascript):

{string} bx = ...;
tuple y{
    string bx;
    _{some other stuff}
}
-snip-
{y} bset[bx];
execute{
    for (var b in bx){
        for (var g in y){
            if (g.bx == b) bset[b].add(g);
        }
    }
}


Now my allocation requires that for each x allocation, the members of {y} in bset be unique.
Or to put it another way:


bx = {"A", "B"}
bset[bx] = [{"AA", "BB", "CC"}, {"DD", "EE", "FF"}

now every y with y.bx == "A" that is allocated to a particular x, must be either AA or BB or CC.
they may not mix.


I'm struggling to represent this.
My first attempt was sort of like this

forall(gt in y, g in y : (g != gt) && (g.bx == gt.bx))){
forall (u in x){
sum(<g, u> in my_dexpr) used[<g, u>] == 0 || sum(<gt, u> in my_dexpr) used[<gt, u>] == 0;
}
}

where used[<y, x>] is obviously the target here.



it sent the number of constraints and variables stratospheric.
it's obviously wrong.
anyone have any hints as to how to do this better?

#DecisionOptimization
#MathematicalProgramming-General

Originally posted by: SystemAdmin

[alan_a said:]

Well it's not much of a change, but this seems to do the job a lot more efficiently.

forall (<gt, ..> in y, <gt1, ..> in y :
(gt != gt1) && (gt.bx == gt1.bx))){
used[<gt, ..>] == 0 || used[<gt1, ..>] == 0;
}


I still can't help thinking there has to be a better way though.
#DecisionOptimization
#MathematicalProgramming-General