Wrap MOI.Complements to Knitro complementarity constraints #130
Conversation
|
Currently, I have tested only with a MOI version of examples/mpec1.jl (btw, let me know if you want me to add this test to MOI). I was planning to test the linear mixed complementarity test this evening! |
Yes, please! I'm considering writing a simple MPEC solver which just does the transformation and passes off to an NLP solver. |
|
@odow I made a PR directly on your development branch. I can write other tests if you think that could prove useful :) |
The more the merrier. |
| # (x_aux complements x) | ||
| # | ||
| function MOI.add_constraint(model::Optimizer, | ||
| func::MOI.VectorAffineFunction, set::MOI.Complements) |
There was a problem hiding this comment.
I don't think you need to support this. You can leave a bridge to convert VectorAffineFunction into VectorOfVariables.
There was a problem hiding this comment.
I agree. However, for some reason, the bridged optimizer is unable to formulate the complementarity constraint in a valid manner. We are calling the function
KN_set_compcons(m::Model, ccTypes::Vector{Cint}, indexComps1::Vector{Cint}, indexComps2::Vector{Cint})with vectors indexComps{1,2} having different length, resulting in an error in Knitro.
There was a problem hiding this comment.
Hmm. That sounds like a bug. Can you make a simple reproducible example?
CC jump-dev/MathOptInterface.jl#913