PySA.jl

PySA Simulated Annealing Interface for JuMP

Installation

julia> import Pkg; Pkg.add("PySA")

julia> using PySA

Getting started

using JuMP
using PySA

model = Model(PySA.Optimizer)
set_silent(model)
set_attribute(model, PySA.NumberOfReads(), 20)
set_attribute(model, PySA.NumberOfSweeps(), 64)
set_attribute(model, PySA.RandomSeed(), 1234)

n = 3
Q = [ -1  2  2
       2 -1  2
       2  2 -1 ]

@variable(model, x[1:n], Bin)
@objective(model, Min, x' * Q * x)

optimize!(model)

for i = 1:result_count(model)
    xi = value.(x; result = i)
    yi = objective_value(model; result = i)
    println("[$i] f($(xi)) = $(yi)")
end

Solver options

PySA.jl exposes the main PySA simulated annealing options as JuMP optimizer attributes. Each option can also be set with MOI.RawOptimizerAttribute using the raw key. Legacy n_* raw keys remain supported for compatibility.

Attribute	Raw key	Alias raw key	Default
PySA.NumberOfSweeps()	num_sweeps	n_sweeps	32
PySA.NumberOfReplicas()	num_replicas	n_replicas	3
PySA.NumberOfReads()	num_reads	n_reads	10
PySA.FinalNumberOfReads()	final_num_reads	-	num_reads
PySA.RandomSeed()	seed	-	nothing
PySA.MinimumTemperature()	min_temp	minimum_temperature	1.0
PySA.MaximumTemperature()	max_temp	maximum_temperature	3.5
PySA.UpdateStrategy()	update_strategy	-	"sequential"
PySA.InitializeStrategy()	initialize_strategy	-	"ones"
PySA.RecomputeEnergy()	recompute_energy	-	false
PySA.SortOutputTemps()	sort_output_temps	-	true
PySA.Parallel()	parallel	-	true

Use set_attribute(model, attribute, value) to override these values before calling optimize!. Use set_silent(model) to disable PySA solver output. When PySA.RandomSeed() is set, PySA.jl runs the backend with parallel=false so NumPy and Numba random streams are reproducible.

Note: The PySA wrapper for Julia is not officially supported by the National Aeronautics and Space Administration. If you are interested in official support for Julia from NASA, let them know!

Note: If you are using PySA.jl in your project, we recommend you to include the .CondaPkg entry in your .gitignore file. The PythonCall module will place a lot of files in this folder when building its Python environment.