Specifying datatype in JuMP expressions and constraints to save memory

Specific Domains Optimization (Mathematical)
,
7

would significantly reduce the number of constraints

There will be the same number of constraints. But you can do something more efficient than constructing the sparse matrix so you can loop over just the non-zero keys.

Perhaps something like

function read_csv()
    data = [Dict{Int,Int}() for k in 1:K]
    open("example_v1.csv", "w") do io
        readline(io)  # skip header
        while !eof(io)
            line = readline(io)
            items = split(line, ' ' )
            (p, w) = parse.(Int, items)
            d = data[w]
            d[p + 1] = get(d, p + 1, 0) + 1
        end
    end
    return data
end

data = read_csv()
model = Model(Gurobi.Optimizer; bridge_constraints = false)
@variable(model, r_i_j[1:I, 1:J], Bin)
@variable(model, y_i_k[1:I, 1:K], Bin)
@constraints(model, begin
    [i in 1:I], sum(y_i_k[i, k] for j in 1:K) <= T3
    [k=1:ceil(Int, K/34)], sum(y_i_k[i, kk] for in 1:I, kk in (1+(k-1)*34):min(k*34, K)) <= 1
    [i=1:I, k=1:K-33], sum(y_i_k[i, k+j] for j in 1:33) <= 1
    [i=1:I, k=1:K], sum(r_i_j[i, j] * y_i_k[i, k] * v for (j, v) in data[k]) >= T1 * y_i_k[i, k]
end)
@objective(model, Min, sum(r_i_j) + (N  - sum(y_i_k)) * p)
1 Like