MiniZinc.jl

MiniZinc.jl

MiniZinc.jl is a wrapper for the MiniZinc constraint modeling language.

It provides a way to write MathOptInterface models to .mzn files, and a way to interact with libminizinc.

Affiliation

This wrapper is maintained by the JuMP community and is not part of the MiniZinc project.

Getting help

If you need help, please ask a question on the JuMP community forum.

If you have a reproducible example of a bug, please open a GitHub issue.

License

MiniZinc.jl is licensed under the MIT License.

The underlying project, MiniZinc/libminizinc, is licensed under the MPL 2.0 license.

Install

Install MiniZinc.jl using the Julia package manager:

import Pkg
Pkg.add("MiniZinc")

In addition to installing the MiniZinc.jl package, this will also download and install the libminizinc binaries. You do not need to install libminizinc separately.

Manual install

Install a copy of libminizinc from minizinc.org or compile one yourself from MiniZinc/libminizinc. Then set the JULIA_LIBMINIZINC_DIR environment variable. For example:

ENV["JULIA_LIBMINIZINC_DIR"] = "C:\\Program Files\\MiniZinc"

Use with MathOptInterface

MiniZinc.jl supports the constraint programming sets defined in MathOptInterface, as well as (in)equality constraints.

The following example solves the following constraint program:

xᵢ ∈ {1, 2, 3} ∀i=1,2,3
zⱼ ∈ {0, 1}    ∀j=1,2
z₁ <-> x₁ != x₂
z₂ <-> x₂ != x₃
z₁ + z₂ = 1

julia> import MiniZinc

julia> import MathOptInterface as MOI

julia> function main()
           model = MOI.Utilities.CachingOptimizer(
               MiniZinc.Model{Int}(),
               MiniZinc.Optimizer{Int}("chuffed"),
           )
           # xᵢ ∈ {1, 2, 3} ∀i=1,2,3
           x = MOI.add_variables(model, 3)
           MOI.add_constraint.(model, x, MOI.Interval(1, 3))
           MOI.add_constraint.(model, x, MOI.Integer())
           # zⱼ ∈ {0, 1}    ∀j=1,2
           z = MOI.add_variables(model, 2)
           MOI.add_constraint.(model, z, MOI.ZeroOne())
           # z₁ <-> x₁ != x₂
           MOI.add_constraint(
               model,
               MOI.VectorOfVariables([z[1], x[1], x[2]]),
               MOI.Reified(MOI.AllDifferent(2)),
           )
           # z₂ <-> x₂ != x₃
           MOI.add_constraint(
               model,
               MOI.VectorOfVariables([z[2], x[2], x[3]]),
               MOI.Reified(MOI.AllDifferent(2)),
           )
           # z₁ + z₂ = 1
           MOI.add_constraint(model, 1 * z[1] + x[2], MOI.EqualTo(1))
           MOI.optimize!(model)
           x_star = MOI.get(model, MOI.VariablePrimal(), x)
           z_star = MOI.get(model, MOI.VariablePrimal(), z)
           return x_star, z_star
       end
main (generic function with 1 method)

julia> main()
([1, 1, 3], [0, 1])

Use with JuMP

You can also call MiniZinc from JuMP, using any solver that libminizinc supports. By default, MiniZinc.jl is compiled with the HiGHS MILP solver, which can be selected by passing the "highs" parameter to MiniZinc.Optimizer:

using JuMP
import MiniZinc
model = Model(() -> MiniZinc.Optimizer{Float64}("highs"))
@variable(model, 1 <= x[1:3] <= 3, Int)
@constraint(model, x in MOI.AllDifferent(3))
@objective(model, Max, sum(i * x[i] for i in 1:3))
optimize!(model)
@show value.(x)

CP-SAT

ORTools_jll currently requires Linux

In order to use the CP-SAT solver from ORTools, use:

import ORTools_jll
path = joinpath(ORTools_jll.artifact_dir, "share", "minizinc", "solvers", "cp-sat.msc")
model = Model(() -> MiniZinc.Optimizer{Float64}(path))

If CP-SAT times out, MiniZinc will terminate the solve without getting the best solution. To fix this, you must set MiniZinc.AllSolutions() to true.

set_time_limit_sec(model, 10)
set_attribute(model, MiniZinc.AllSolutions(), true)

MathOptInterface API

The MiniZinc Optimizer{T} supports the following constraints and attributes.

List of supported objective functions:

• MOI.ObjectiveFunction{MOI.ScalarAffineFunction{T}}
• MOI.ObjectiveFunction{MOI.ScalarQuadraticFunction{T}}
• MOI.ObjectiveFunction{MOI.VariableIndex}

List of supported variable types:

• MOI.Reals

List of supported constraint types:

• MOI.ScalarAffineFunction{T} in MOI.EqualTo{T}
• MOI.ScalarAffineFunction{T} in MOI.GreaterThan{T}
• MOI.ScalarAffineFunction{T} in MOI.Integer
• MOI.ScalarAffineFunction{T} in MOI.Interval{T}
• MOI.ScalarAffineFunction{T} in MOI.LessThan{T}
• MOI.ScalarAffineFunction{T} in MOI.ZeroOne
• MOI.VariableIndex in MOI.EqualTo{T}
• MOI.VariableIndex in MOI.GreaterThan{T}
• MOI.VariableIndex in MOI.Integer
• MOI.VariableIndex in MOI.Interval{T}
• MOI.VariableIndex in MOI.LessThan{T}
• MOI.VariableIndex in MOI.Parameter{T}
• MOI.VariableIndex in MOI.Semicontinuous{T}
• MOI.VariableIndex in MOI.Semiinteger{T}
• MOI.VariableIndex in MOI.ZeroOne
• MOI.VectorOfVariables in MOI.AllDifferent
• MOI.VectorOfVariables in MOI.BinPacking{T}
• MOI.VectorOfVariables in MOI.Circuit
• MOI.VectorOfVariables in MOI.CountAtLeast
• MOI.VectorOfVariables in MOI.CountBelongs
• MOI.VectorOfVariables in MOI.CountDistinct
• MOI.VectorOfVariables in MOI.CountGreaterThan
• MOI.VectorOfVariables in MOI.Cumulative
• MOI.VectorOfVariables in MOI.Path
• MOI.VectorOfVariables in MOI.Table{T}

List of supported model attributes:

• MOI.ConflictStatus()
• MOI.NLPBlock()
• MOI.Name()
• MOI.ObjectiveBound()
• MOI.ObjectiveSense()

List of supported constraint attributes:

• MOI.ConstraintConflictStatus()

Conflicts (IIS)

For an infeasible model, MOI.compute_conflict! finds a minimal conflicting subset of constraints (an Irreducible Inconsistent Subsystem), after which MOI.ConstraintConflictStatus reports which constraints participate. For example, this model is infeasible because x[1] + x[2] cannot be both >= 18 and <= 5:

model = MOI.Utilities.CachingOptimizer(
    MiniZinc.Model{Int}(),
    MiniZinc.Optimizer{Int}("chuffed"),
)
x = MOI.add_variables(model, 2)
MOI.add_constraint.(model, x, MOI.Interval(1, 10))
f = 1 * x[1] + 1 * x[2]
MOI.add_constraint(model, f, MOI.GreaterThan(18))  # x[1] + x[2] >= 18
MOI.add_constraint(model, f, MOI.LessThan(5))       # x[1] + x[2] <= 5
MOI.optimize!(model)
if MOI.get(model, MOI.TerminationStatus()) == MOI.INFEASIBLE
    MOI.compute_conflict!(model)
    if MOI.get(model, MOI.ConflictStatus()) == MOI.CONFLICT_FOUND
        # Query each constraint you added to `model` for its participation.
        # Variable bounds always read NOT_IN_CONFLICT (see below).
        for (F, S) in MOI.get(model, MOI.ListOfConstraintTypesPresent())
            for ci in MOI.get(model, MOI.ListOfConstraintIndices{F,S}())
                status = MOI.get(model, MOI.ConstraintConflictStatus(), ci)
                # status is MOI.IN_CONFLICT or MOI.NOT_IN_CONFLICT
            end
        end
    end
end

This uses findMUS, which is not part of the standard MiniZinc distribution; it is supplied by the FindMUS_jll dependency, so no extra setup is required.

Conflicts cover modeling constraints only; variable bounds are folded into the variable declarations and never appear in a conflict.

Conflict analysis always uses the Chuffed subsolver, regardless of the solver passed to MiniZinc.Optimizer, so a model outside Chuffed’s support (for example, one with floating-point variables) reports NO_CONFLICT_FOUND.

Options

Set options using MOI.RawOptimizerAttribute in MOI or set_attribute in JuMP.

MiniZinc.jl supports the following options:

• model_filename::String = "": the location at which to write out the .mzn file during optimization. This option can be helpful during debugging. If left empty, a temporary file will be used instead.

• MiniZinc.CmdFlags(): set this option to pass extra flags to the minizinc command, such as "-p 4 --free-search". Flags that MiniZinc.jl sets itself, like --num-solutions and --time-limit, should be set with their options instead.

• MOI.SolutionLimit: set this option to a positive integer to return up to the limit number of solutions.

• MiniZinc.AllSolutions(): set this option to true to return all solutions of a satisfaction problem (up to MOI.SolutionLimit), or every improving solution of an optimization problem. This is the -a flag of MiniZinc.