How to define new variables into the model with a while loop using JuMP and Gurobi

[MirelEce]

I am new to Julia and Gurobi and trying to add a new variable and two constraints associated with the newly added variable to my model in each iteration in the while loop. I am not sure how to parametrically denote new variables in this case. Consider the following simple example:

k=3
yn=[5, 6, 8]
yp=[12, 65, 46]
epsilon=0.0001

m = Model(solver = GurobiSolver())
@variable(m, w[1:k] >= 0)
@variable(m, beta >= 0)
@variable(m, t >= 0)
@objective(m, Min, t)
@constraint(m, sum{w[i], i in 1:k}==1)
@constraint(m,sum{w[i]*yn[i], i in 1:kk} >= sum{w[i]*yp[i], i in 1:kk} + beta)
@constraint(m, t >= beta)
@constraint(m, beta >= epsilon)

wf=[0.2,0.5,0.3]
eps=0.001
i=1
while flag == 0
    @variable(m, z, Bin)  #Here I actually want to create a different variable zi (z1,z2,z3... as it iterates) 
    @constraint(m, w[1]-wf[1]<= -1*eps+(1+eps)*z) #Then there will be zi here as well
    @constraint(m, w[1]-wf[1]>= eps-(1-z)*(1+eps)) #It will be zi here too
    statusm=solve(m)

    if string(statusm)=="Optimal"
        wf = getvalue(w)
        println(wf)
    elseif string(statusfm)=="Infeasible"
        flag=1
    end
end

Here are the options I tried:

1- leaving it as such -> I am afraid that as the variables' names are not unique, the model would just treat all variables as the same, i.e., all variables denoted as z would be the same.

2- trying to define the variables anonymously by doing: z= @variable(m, binary=true) -> after writing down the model by print(m), I still see that all of the variables are anon, so it does not differentiate between them.

3- Defining a variable as an array z, but I do not know the size of the array in advance, as it will continue to grow until the while loop condition is no longer satisfied (which is when the model I am trying to solve becomes infeasible) and defining a variable of an array whose size is 10^30 is highly inefficient- may not even work as the setup could not be completed..

After searching online, maybe I can use dictionary to solve my problem; however, I am not sure how to do so in my case. I would truly appreciate it if someone could help me with this. Thanks!

[odow]

2. is the correct way to go. The names are printed as anon, but the variables are different.

Here is the documentation: https://jump.dev/JuMP.jl/v0.21.1/variables/#Anonymous-JuMP-variables-1

p.s., in the future, please post questions like this on the forum: https://discourse.julialang.org/c/domain/opt/13. We try to keep Github issues for bug reports specific to each package.

[MirelEce]

Thank you so much for your response. If I use option 2, is there also a way that I can get the optimal values of the anonymous variables somehow? If I use getvalue(z), it only gives me the optimal value of the latest _anon, but not the others.

About the forum- I didn't know, my bad! Will do next time. Feel free to delete this post afterwards to avoid any confusion.

[odow]

Do something like:

z_vec = []
while true
    z = @variable(model)
    push!(z_vec, z)
    # optimize
    value(z_vec[end])
end

Note that you appear to be using an old version of JuMP. You should upgrade to the most recent version.

[MirelEce]

Thank you so much!