I just change to JuMP 0.19 and I get some problem with my code.

I construct a big JuMP model with a lot of specific variables and constraints, and I want to extract constraint matrix and vectors from the model constructed with JuMP:

Min c*x
s.t. l <= Ax <= u
lb <= x <= ub

I write, 2 years ago the following piece of code, I don’t know if it was optimal, but it worked

JuMP.build(model)
c = MathProgBase.getobj(JuMP.internalmodel(model))
A = MathProgBase.getconstrmatrix(JuMP.internalmodel(model))
lb = MathProgBase.getvarLB(JuMP.internalmodel(model))
ub = MathProgBase.getvarUB(JuMP.internalmodel(model))
l = MathProgBase.getconstrLB(JuMP.internalmodel(model))
u = MathProgBase.getconstrUB(JuMP.internalmodel(model))

But know, with JuMP 0.19 and MathOptInterface, it doesn’t work anymore and I didn’t find anything in documentations or any topics related. Can anybody indicate me where to find information or what is the best thing to do?

I don’t know the matrix A and vector l,u,lb,ub and c. I use JuMP to create all my variables and constraints (I describe with them a unit commitment problem for hydro power plants) , then I used the code I wrote previously to get these information (A,…etc). The problem is to complicated to construct “by-hand” constraints matrices.

I need this because I construct these matrices A for several power plants with the same code (with different parameters) and then link all power plant by using all these A matrices and vectors.

If a have 2 power plants, my problem will become

Min c1x1+c2x2

s.t. [ l1 l2 ] <= [A1 0
0 A2] * [x1 x2] <= [u1 u2]
+ some constraints that link the two power plant depending on x1 and x2

If there exist a way to merge such optimization problem directly with JuMP and MathOptInterface I will be happy.

See Constraints · JuMP. It requires a bit more code than before because JuMP 0.19 supports a lot of new constraints and has no specific concept of a constraint matrix anymore.

You should, think carefully about what you what and whats the best way to achieve that, sometimes getting the matrices is a simple but not optimal method. I recommend thinking twice before extracting those matrices.

Can’t you just write this directly in one go? With the plant as an index in variables and constraints?

P = 3 # Number of plants
U = [2, 2, 4] # Number of units at each plant
model = Model()
@variable(model, unit_commit[p=1:P, u=1:U[p]], Bin)
@objective(model, Min, sum(
(p + u) * unit_commit[p, u] for p=1:P for u=1:U[p]))
@constraint(model, [p=1:P, i=2:P[p]], unit_commit[p, i-1] <= unit_commit[p, i])
@constraint(model, sum(unit_commit) >= 4)