Dualizing ParametricOptInterface

[matbesancon]

A model created with ParameterJuMP does not currently suppport dualization, the parameters are lost in the way. Given the form given in the manual, a parameter p such that one constraint is of the form A_i x + b_i + p in C_i ends up with an additive coefficient in the objective of the dual: y_i (b_i + p).

My intuition is that things would be smoother with ParameterJuMP.jl implemented at the MOI level and with the sets being for example MOI.EqualTo{ParamAffineExpression}, but I recognize it puts a burden on maintenance efforts.

As of now, I haven't figured out a place where the parameters appear in types of the JuMP model or constraints so it might be tricky without modifying ParameterJuMP.jl

[joaquimg]

The current plan is that we can connect with ParametricOptInterface, once that is stable enough.

[dsigler1234]

Hello,

I was wondering if Dualization.jl supports ParametricOptInterface parameters for LPs at this point or if this is still work in progress?

[guilhermebodin]

Hi @dsigler1234 I think it still does not work but we haven`t recently tested it. I might look at it later this week but would not count on it working.

[dsigler1234]

@guilhermebodin thanks for your quick reply. Would a pull request with this capability for LPs be of interest?