RoundAndSwap.jl
| Documentation | Build Status |
|---|---|
  |
  |
RoundAndSwap.jl is a library which implements the Round and Swap algorithm, this implementation was inspired by Filippo Pecci's usage in Bounds and Convex Heuristics for Bi-Objective Sensor Placement in Water Networks.
Getting started
- Create your JuMP model, and optimize
# Define how many variables you want fixed
num_to_fix = 2
using JuMP, HiGHS
model = Model(HiGHS.Optimizer)
@variable(model, 0 ≤ a ≤ 1)
@variable(model, 0 ≤ b ≤ 1)
@variable(model, 0 ≤ c ≤ 1)
@variable(model, 0 ≤ d ≤ 1)
@constraint(model, a + b + c + d == num_to_fix)
@objective(model, Max, (a + b) + (2 * (b + c)) + (3 * (c - d)) + (4 * (d + a)))
# To demonstate functionality we will fix b and d
fix(b, 0.8; force=true)
fix(d, 0.8; force=true)
optimize!(model)- Identify which variable we are considering making
1
consider_swapping = [a,b,c,d]- Round variables closest to
1, this will be b and d as we fixed them at0.8.
using RoundAndSwap
round!(consider_swapping, num_to_fix)- Begin swapping
best_swap, swapper = swap(model, consider_swapping)- Review your best swap
julia> best_swap
1-element Vector{Swap}:
Swap
existing: b
new: a
obj_value: 10.0
success: true
all_fixed: [:a, :c]
termination_status: OPTIMAL
solve_time: 0.0004438320000872409
swap_number: 7As you can see in this case we found the globally optimal solution of 10. Note: Round and Swap does not provide guarantees of global optimility.
caveats
- Currently can only set variables to
0or1.
Performance
In testing I've done 536 swaps with round_and_swap taking 174.1 seconds with 172.8 seconds of that being in the optimizer, giving RoundAndSwap.jl an overhead of 0.75%.