Ordering of OptimizationVariables in an OptimizationProblem

Matt J

2021 5 24

0 个回答

5 次查看（30 天）

1 个投票

When I use the problem-based optimization framework to create an OptimizationProblem, e.g.,

r=optimvar('r',5,'LowerBound',0,'UpperBound',1);
B= optimvar('B', 3);
prob=optimproblem('Constraints', sum(r)<=1,'Objective', sum(r)+sum(B));

the ordering of the variables in prob appears to be alphabetical,

prob.Variables
ans = struct with fields:
    B: [3×1 optim.problemdef.OptimizationVariable]
    r: [5×1 optim.problemdef.OptimizationVariable]

However, I would prefer the ordering to be r first, then B. I thought I might be able to use orderfields() to change it, but no luck:

prob.Variables=orderfields(prob.Variables,{'r','B'})
OptimizationProblem property 'Variables' is read-only.

Is there a supported way of doing this?

The reason I want to do this is so that when I use,

s=prob2struct(prob);

the sub-matrix partitions of s.lb, s.ub, s.Aeq, etc... corresponding to r and B will have an ordering that I can know and control in advance. I'm assuming this ordering is dictated by the ordering of the fields in prob.Variables.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Matt J 2021-5-25

编辑：Matt J 2021-5-25

@Bjorn Gustavsson

IMO, it is at minimum often a convenient tool for setting up linear constraints. It spares you the mental gymnastics of having to figure out the matrix form of the constraints, especially when the constraints may involve multi-dimensional sums. Also, when the original mathematical formulation of the problem involves multiple unknown arrays like r and B, it spares you the effort of having to concatenate them into a single unknown vector.