Add support for vector-valued objectives

docs/Project.toml

@@ -12,6 +12,7 @@ JSONSchema = "7d188eb4-7ad8-530c-ae41-71a32a6d4692"
 JuMP = "4076af6c-e467-56ae-b986-b466b2749572"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 MathOptInterface = "b8f27783-ece8-5eb3-8dc8-9495eed66fee"
+MultiObjectiveAlgorithms = "0327d340-17cd-11ea-3e99-2fd5d98cecda"
 Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
@@ -33,7 +34,8 @@ Ipopt = "=1.1.0"
 JSON = "0.21"
 JSONSchema = "1"
 Literate = "2.8"
-MathOptInterface = "=1.11.5"
+MathOptInterface = "=1.12.0"
+MultiObjectiveAlgorithms = "=0.1.1"
 Plots = "1"
 SCS = "=1.1.3"
 SQLite = "1"

docs/make.jl

@@ -131,6 +131,8 @@ const _PAGES = [
             "tutorials/linear/factory_schedule.md",
             "tutorials/linear/finance.md",
             "tutorials/linear/geographic_clustering.md",
+            "tutorials/linear/multi_objective_knapsack.md",
+            "tutorials/linear/multi_objective_examples.md",
             "tutorials/linear/knapsack.md",
             "tutorials/linear/multi.md",
             "tutorials/linear/n-queens.md",
@@ -284,7 +286,7 @@ function _add_moi_pages()
     !!! warning
         This documentation in this section is a copy of the official
         MathOptInterface documentation available at
-        [https://jump.dev/MathOptInterface.jl/v1.11.5](https://jump.dev/MathOptInterface.jl/v1.11.5).
+        [https://jump.dev/MathOptInterface.jl/v1.12.0](https://jump.dev/MathOptInterface.jl/v1.20.0).
         It is included here to make it easier to link concepts between JuMP and
         MathOptInterface.
     """

docs/src/manual/objective.md

@@ -157,3 +157,84 @@ julia> @objective(model, Min, 2x)
 julia> @objective(model, Max, objective_function(model))
 2 x
 ```
+
+## Set a vector-valued objective
+
+Define a multi-objective optimization problem by passing a vector of objectives:
+
+```jldoctest; setup = :(model=Model())
+julia> @variable(model, x[1:2]);
+
+julia> @objective(model, Min, [1 + x[1], 2 * x[2]])
+2-element Vector{AffExpr}:
+ x[1] + 1
+ 2 x[2]
+
+julia> f = objective_function(model)
+2-element Vector{AffExpr}:
+ x[1] + 1
+ 2 x[2]
+```
+
+!!! tip
+    The [Multi-objective knapsack](@ref) tutorial provides an example of
+    solving a multi-objective integer program.
+
+In most cases, multi-objective optimization solvers will return multiple
+solutions, corresponding to points on the Pareto frontier. See [Multiple solutions](@ref)
+for information on how to query and work with multiple solutions.
+
+Note that you must set a single objective sense, that is, you cannot have
+both minimization and maximization objectives. Work around this limitation by
+choosing `Min` and negating any objectives you want to maximize:
+
+```jldoctest; setup = :(model=Model())
+julia> @variable(model, x[1:2]);
+
+julia> @expression(model, obj1, 1 + x[1])
+x[1] + 1
+
+julia> @expression(model, obj2, 2 * x[1])
+2 x[1]
+
+julia> @objective(model, Min, [obj1, -obj2])
+2-element Vector{AffExpr}:
+ x[1] + 1
+ -2 x[1]
+```
+
+Defining your objectives as expressions allows flexibility in how you can solve
+variations of the same problem, with some objectives removed and constrained to
+be no worse that a fixed value.
+
+```jldoctest; setup = :(model=Model())
+julia> @variable(model, x[1:2]);
+
+julia> @expression(model, obj1, 1 + x[1])
+x[1] + 1
+
+julia> @expression(model, obj2, 2 * x[1])
+2 x[1]
+
+julia> @expression(model, obj3, x[1] + x[2])
+x[1] + x[2]
+
+julia> @objective(model, Min, [obj1, obj2, obj3])  # Three-objective problem
+3-element Vector{AffExpr}:
+ x[1] + 1
+ 2 x[1]
+ x[1] + x[2]
+
+julia> # optimize!(model), look at the solution, talk to stakeholders, then
+       # decide you want to solve a new problem where the third objective is
+       # removed and constrained to be better than 2.0.
+       nothing
+
+julia> @objective(model, Min, [obj1, obj2])   # Two-objective problem
+2-element Vector{AffExpr}:
+ x[1] + 1
+ 2 x[1]
+
+julia> @constraint(model, obj3 <= 2.0)
+x[1] + x[2] ≤ 2.0
+```

docs/src/manual/solutions.md

@@ -573,6 +573,10 @@ for i in 2:result_count(model)
 end
 ```
 
+!!! tip
+    The [Multi-objective knapsack](@ref) tutorial provides an example of
+    querying multiple solutions.
+
 ## Checking feasibility of solutions
 
 To check the feasibility of a primal solution, use

docs/src/tutorials/linear/multi_objective_examples.jl

@@ -0,0 +1,124 @@
+# Copyright 2017, Iain Dunning, Joey Huchette, Miles Lubin, and contributors    #src
+# This Source Code Form is subject to the terms of the Mozilla Public License   #src
+# v.2.0. If a copy of the MPL was not distributed with this file, You can       #src
+# obtain one at https://mozilla.org/MPL/2.0/.                                   #src
+
+# # Simple multi-objective examples
+
+# This tutorial contains a number of examples of multi-objective programs from
+# the literature.
+
+# ## Required packages
+
+# This tutorial requires the following packages:
+
+using JuMP
+import HiGHS
+import MultiObjectiveAlgorithms as MOA
+
+# ## Bi-objective linear problem
+
+# This is example is taken from Example 6.3 (from Steuer, 1985), page 154 of
+# Multicriteria Optimization (2nd ed), M. Ehrgott, Springer 2005. The code was
+# adapted from an example in [vOptGeneric](https://github.com/vOptSolver/vOptGeneric.jl)
+# by [@xgandibleux](https://github.com/xgandibleux).
+
+model = Model()
+set_silent(model)
+@variable(model, x1 >= 0)
+@variable(model, 0 <= x2 <= 3)
+@objective(model, Min, [3x1 + x2, -x1 - 2x2])
+@constraint(model, 3x1 - x2 <= 6)
+set_optimizer(model, () -> MOA.Optimizer(HiGHS.Optimizer))
+set_optimizer_attribute(
+    model,
+    MOA.Algorithm(),
+    MOA.Lexicographic(; all_permutations = true),
+)
+optimize!(model)
+solution_summary(model)
+
+#-
+
+for i in 1:result_count(model)
+    print(i, ": z = ", round.(Int, objective_value(model; result = i)), " | ")
+    println("x = ", value.([x1, x2]; result = i))
+end
+
+# ## Bi-objective linear assignment problem
+
+# This is example is taken from Example 9.38 (from Ulungu and Teghem, 1994),
+# page 255 of Multicriteria Optimization (2nd ed), M. Ehrgott, Springer 2005.
+# The code was adapted from an example in [vOptGeneric](https://github.com/vOptSolver/vOptGeneric.jl)
+# by [@xgandibleux](https://github.com/xgandibleux).
+
+C1 = [5 1 4 7; 6 2 2 6; 2 8 4 4; 3 5 7 1]
+C2 = [3 6 4 2; 1 3 8 3; 5 2 2 3; 4 2 3 5]
+n = size(C2, 1)
+model = Model()
+set_silent(model)
+@variable(model, x[1:n, 1:n], Bin)
+@objective(model, Min, [sum(C1 .* x), sum(C2 .* x)])
+@constraint(model, [i = 1:n], sum(x[i, :]) == 1)
+@constraint(model, [j = 1:n], sum(x[:, j]) == 1)
+set_optimizer(model, () -> MOA.Optimizer(HiGHS.Optimizer))
+set_optimizer_attribute(model, MOA.Algorithm(), MOA.EpsilonConstraint())
+optimize!(model)
+solution_summary(model)
+
+#-
+
+for i in 1:result_count(model)
+    print(i, ": z = ", round.(Int, objective_value(model; result = i)), " | ")
+    println("x = ", round.(Int, value.(x; result = i)))
+end
+
+# ## Bi-objective shortest path problem
+
+# This is example is taken from Exercise 9.5 page 269 of Multicriteria
+# Optimization (2nd edt), M. Ehrgott, Springer 2005. The code was adapted from
+# an example in [vOptGeneric](https://github.com/vOptSolver/vOptGeneric.jl) by
+# [@xgandibleux](https://github.com/xgandibleux).
+
+M = 50
+C1 = [
+    M 4 5 M M M
+    M M 2 1 2 7
+    M M M 5 2 M
+    M M 5 M M 3
+    M M M M M 4
+    M M M M M M
+]
+C2 = [
+    M 3 1 M M M
+    M M 1 4 2 2
+    M M M 1 7 M
+    M M 1 M M 2
+    M M M M M 2
+    M M M M M M
+]
+n = size(C2, 1)
+model = Model()
+set_silent(model)
+@variable(model, x[1:n, 1:n], Bin)
+@objective(model, Min, [sum(C1 .* x), sum(C2 .* x)])
+@constraint(model, sum(x[1, :]) == 1)
+@constraint(model, sum(x[:, n]) == 1)
+@constraint(model, [i = 2:n-1], sum(x[i, :]) - sum(x[:, i]) == 0)
+set_optimizer(model, () -> MOA.Optimizer(HiGHS.Optimizer))
+set_optimizer_attribute(model, MOA.Algorithm(), MOA.EpsilonConstraint())
+optimize!(model)
+solution_summary(model)
+
+#-
+
+for i in 1:result_count(model)
+    print(i, ": z = ", round.(Int, objective_value(model; result = i)), " | ")
+    X = round.(Int, value.(x; result = i))
+    print("Path:")
+    for ind in findall(val -> val ≈ 1, X)
+        i, j = ind.I
+        print(" $i->$j")
+    end
+    println()
+end