MOI tests: intlinear (#533)

* add intlinear tests * delete UNKNOWN_TERMINATION_STATUS and fix user_algorithms_tests
atoptima · May 24, 2021 · c6aba41 · c6aba41
1 parent 5c36414
commit c6aba41
Show file tree

Hide file tree

Showing 8 changed files with 107 additions and 98 deletions.
diff --git a/src/Algorithm/basic/solveipform.jl b/src/Algorithm/basic/solveipform.jl
@@ -61,7 +61,6 @@ function run!(algo::SolveIpForm, env::Env, data::ModelData, input::OptimizationI
     if !ip_supported
         @warn "Optimizer of formulation with id =", getuid(form),
               " does not support integer variables. Skip SolveIpForm algorithm."
-        setterminationstatus!(result, UNKNOWN_TERMINATION_STATUS)
         return OptimizationOutput(result)
     end
 

diff --git a/src/Algorithm/utilities/optimizationstate.jl b/src/Algorithm/utilities/optimizationstate.jl
@@ -36,7 +36,7 @@ end
 """
     OptimizationState(
         form; 
-        termination_status = UNKNOWN_TERMINATION_STATUS,
+        termination_status = OPTIMIZE_NOT_CALLED,
         ip_primal_bound = nothing, 
         ip_dual_bound = nothing, 
         lp_primal_bound = nothing, 
@@ -49,9 +49,9 @@ end
         insert_function_lp_dual_sols = bestbound!
         )
 
-A convenient structure to maintain and return solutions and bounds of a formulation `form` during an
-optimization process. The termination statuses is considered as
-unknown by default. You can define the initial incumbent bounds using `ip_primal_bound`,
+A convenient structure to maintain and return solutions and bounds of a formulation
+`form` during an optimization process. The termination status is `OPTIMIZE_NOT_CALLED`
+by default. You can define the initial incumbent bounds using `ip_primal_bound`,
 `ip_dual_bound`, `lp_primal_bound`, and `lp_primal_bound` keyword arguments. Incumbent
 bounds are set to infinite (according to formulation objective sense) by default.
 You can store three types of solutions `ip_primal_sols`, `lp_primal_sols`, and `lp_dual_sols`.
@@ -64,7 +64,7 @@ best bound.
 """
 function OptimizationState(
     form::F;
-    termination_status::TerminationStatus = UNKNOWN_TERMINATION_STATUS,
+    termination_status::TerminationStatus = OPTIMIZE_NOT_CALLED,
     ip_primal_bound = nothing,
     ip_dual_bound = nothing,
     lp_primal_bound = nothing,

diff --git a/src/ColunaBase/ColunaBase.jl b/src/ColunaBase/ColunaBase.jl
@@ -19,8 +19,8 @@ export Bound, Solution, getvalue, isbetter, diff, gap, printbounds, getsol,
     getstatus, remove_until_last_point
 
 # Statuses
-export TerminationStatus, SolutionStatus, OPTIMAL, INFEASIBLE, TIME_LIMIT, 
-    NODE_LIMIT, OTHER_LIMIT, UNKNOWN_TERMINATION_STATUS, UNCOVERED_TERMINATION_STATUS, 
+export TerminationStatus, SolutionStatus, OPTIMIZE_NOT_CALLED, OPTIMAL,
+    INFEASIBLE, TIME_LIMIT, NODE_LIMIT, OTHER_LIMIT, UNCOVERED_TERMINATION_STATUS, 
     FEASIBLE_SOL, INFEASIBLE_SOL, UNKNOWN_FEASIBILITY, UNKNOWN_SOLUTION_STATUS, 
     UNCOVERED_SOLUTION_STATUS, convert_status
 

diff --git a/src/ColunaBase/solsandbounds.jl b/src/ColunaBase/solsandbounds.jl
@@ -133,24 +133,24 @@ MOI [`TerminationStatusCode`](https://jump.dev/MathOptInterface.jl/stable/apiref
 is translated into a Coluna `TerminationStatus`.
 
 Description of the termination statuses: 
-- `OPTIMAL` : the algorithm found a global optimal solution given the optimality tolerance.
+- `OPTIMAL` : the algorithm found a global optimal solution given the optimality tolerance
 - `INFEASIBLE` : the algorithm proved infeasibility
 - `TIME_LIMIT` : the algorithm stopped because of the time limit
 - `NODE_LIMIT` : the branch-and-bound based algorithm stopped due to the node limit
 - `OTHER_LIMIT` : the algorithm stopped because of a limit that is neither the time limit 
 nor the node limit
 
 If the algorithm has not been called, the default value of the termination status should be:
-- `UNKNOWN_TERMINATION_STATUS`
+- `OPTIMIZE_NOT_CALLED`
 
 If the subsolver called through MOI returns a 
-`TerminationStatusCode` that is not `MOI.OPTIMAL`, `MOI.INFEASIBLE`, `MOI.TIME_LIMIT`, `MOI.NODE_LIMIT`, or 
-`MOI.OTHER_LIMIT`:
+`TerminationStatusCode` that is not `MOI.OPTIMIZE_NOT_CALLED`, `MOI.OPTIMAL`, `MOI.INFEASIBLE`,
+`MOI.TIME_LIMIT`, `MOI.NODE_LIMIT`, or `MOI.OTHER_LIMIT`:
 - `UNCOVERED_TERMINATION_STATUS` : should not be used by a Coluna algorithm
 """
 @enum(
-    TerminationStatus, OPTIMAL, INFEASIBLE, TIME_LIMIT, NODE_LIMIT, OTHER_LIMIT, 
-    UNKNOWN_TERMINATION_STATUS, UNCOVERED_TERMINATION_STATUS
+    TerminationStatus, OPTIMIZE_NOT_CALLED, OPTIMAL, INFEASIBLE,
+    TIME_LIMIT, NODE_LIMIT, OTHER_LIMIT, UNCOVERED_TERMINATION_STATUS
 )
 
 """
@@ -162,7 +162,7 @@ Description of the solution statuses:
 
 If there is no solution or if we don't have information about the solution, the 
 solution status should be :
-- `UNKWNOW_SOLUTION_STATUS`
+- `UNKNOWN_SOLUTION_STATUS`
 
 """
 @enum(
@@ -180,6 +180,7 @@ Convert a termination or solution `status` of a given type to the corresponding
 This method is used to communicate between Coluna and MathOptInterface.
 """
 function convert_status(moi_status::MOI.TerminationStatusCode)
+    moi_status == MOI.OPTIMIZE_NOT_CALLED && return OPTIMIZE_NOT_CALLED
     moi_status == MOI.OPTIMAL && return OPTIMAL
     moi_status == MOI.INFEASIBLE && return INFEASIBLE
     moi_status == MOI.TIME_LIMIT && return TIME_LIMIT
@@ -189,6 +190,7 @@ function convert_status(moi_status::MOI.TerminationStatusCode)
 end
 
 function convert_status(coluna_status::TerminationStatus)
+    coluna_status == OPTIMIZE_NOT_CALLED && return MOI.OPTIMIZE_NOT_CALLED
     coluna_status == OPTIMAL && return MOI.OPTIMAL
     coluna_status == INFEASIBLE && return MOI.INFEASIBLE
     coluna_status == TIME_LIMIT && return MOI.TIME_LIMIT

diff --git a/src/MOIwrapper.jl b/src/MOIwrapper.jl
@@ -28,7 +28,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
     constrs_on_single_var_to_vars::Dict{MOI.ConstraintIndex, VarId}
     constrs_on_single_var_to_names::Dict{MOI.ConstraintIndex, String}
     names_to_constrs::Dict{String, MOI.ConstraintIndex}
-    result::Union{Nothing,OptimizationState}
+    result::OptimizationState
     default_optimizer_builder::Union{Nothing, Function}
 
     feasibility_sense::Bool # Coluna supports only Max or Min.
@@ -47,6 +47,7 @@ mutable struct Optimizer <: MOI.AbstractOptimizer
         model.constrs_on_single_var_to_vars = Dict{MOI.ConstraintIndex, VarId}()
         model.constrs_on_single_var_to_names = Dict{MOI.ConstraintIndex, String}()
         model.names_to_constrs = Dict{String, MOI.ConstraintIndex}()
+        model.result = OptimizationState(get_optimization_target(model.inner))
         model.default_optimizer_builder = nothing
         model.feasibility_sense = false
         return model
@@ -587,6 +588,7 @@ function MOI.empty!(model::Coluna.Optimizer)
     if model.default_optimizer_builder !== nothing
         set_default_optimizer_builder!(model.inner, model.default_optimizer_builder)
     end
+    model.result = OptimizationState(get_optimization_target(model.inner))
     return
 end
 

diff --git a/src/MathProg/MOIinterface.jl b/src/MathProg/MOIinterface.jl
@@ -100,7 +100,9 @@ function enforce_kind_in_optimizer!(form::Formulation, v::Variable)
     moirecord = getmoirecord(v)
     moi_kind = getkind(moirecord)
     if moi_kind.value != -1
-        MOI.delete(inner, moi_kind)
+        if MOI.is_valid(inner, moi_kind)
+            MOI.delete(inner, moi_kind)
+        end
         setkind!(moirecord, MoiVarKind())
     end
     if kind != Continuous # Continuous is translated as no constraint in MOI

diff --git a/test/MathOptInterface/MOI_wrapper.jl b/test/MathOptInterface/MOI_wrapper.jl
@@ -10,84 +10,12 @@ const CONFIG = MOIT.TestConfig(atol=1e-6, rtol=1e-6)
     @test MOI.get(OPTIMIZER, MOI.SolverName()) == "Coluna"
 end
 
-@testset "Kpis" begin
-    data = CLD.GeneralizedAssignment.data("smallgap3.txt")
-
-    coluna = JuMP.optimizer_with_attributes(
-        CL.Optimizer,
-        "params" => CL.Params(
-            solver = ClA.TreeSearchAlgorithm(
-                conqueralg = ClA.ColCutGenConquer(
-                    colgen = ClA.ColumnGeneration(max_nb_iterations = 8)
-                ), maxnumnodes = 4
-            )
-        ),
-        "default_optimizer" => GLPK.Optimizer
-    )
-
-    problem, x, dec = CLD.GeneralizedAssignment.model(data, coluna)
-
-    JuMP.optimize!(problem)
-
-    @test MOI.get(problem, MOI.NodeCount()) == 4
-    @test isa(MOI.get(problem, MOI.SolveTime()), Float64)
-end
-
 @testset "supports_default_copy_to" begin
     @test MOIU.supports_default_copy_to(OPTIMIZER, false)
     # Use `@test !...` if names are not supported
     @test MOIU.supports_default_copy_to(OPTIMIZER, true)
 end
 
-@testset "branching_priority" begin
-    coluna = JuMP.optimizer_with_attributes(
-        Coluna.Optimizer,
-        "params" => CL.Params(solver=ClA.SolveIpForm()),
-        "default_optimizer" => GLPK.Optimizer
-    )
-    model = BlockModel(coluna, direct_model=true)
-    @variable(model, x)
-
-    @test BlockDecomposition.branchingpriority(model, x) == 1
-    BlockDecomposition.branchingpriority!(model, x, 2)
-    @test BlockDecomposition.branchingpriority(model, x) == 2
-
-    model2 = BlockModel(coluna)
-    @variable(model2, x)
-
-    @test BlockDecomposition.branchingpriority(model2, x) == 1
-    BlockDecomposition.branchingpriority!(model2, x, 2)
-    @test BlockDecomposition.branchingpriority(model2, x) == 2
-end
-
-@testset "write_to_file" begin
-    coluna = JuMP.optimizer_with_attributes(
-        Coluna.Optimizer,
-        "params" => CL.Params(solver=ClA.SolveIpForm()),
-        "default_optimizer" => GLPK.Optimizer
-    )
-    model = BlockModel(coluna, direct_model=true)
-    @variable(model, x)
-
-    @constraint(model, x <= 1)
-    @objective(model, Max, x)
-    optimize!(model)
-
-    dest = MOI.FileFormats.Model(format = MOI.FileFormats.FORMAT_MPS)
-    MOI.copy_to(dest, model)
-    MOI.write_to_file(dest, "model.mps")
-
-    filedata = MOI.FileFormats.Model(format = MOI.FileFormats.FORMAT_MPS)
-    MOI.read_from_file(filedata, "model.mps")
-
-    model2 = BlockModel(coluna, direct_model=true)
-    MOI.copy_to(model2, filedata)
-    optimize!(model2)
-
-    # the model is always written as a minimization problem
-    @test JuMP.objective_value(model) == -JuMP.objective_value(model2)
-end
-
 const UNSUPPORTED_TESTS = [
     "solve_qcp_edge_cases", # Quadratic constraints not supported
     "delete_nonnegative_variables", # variable deletion not supported
@@ -161,13 +89,18 @@ const LP_TESTS = [
     MOIT.unittest(OPTIMIZER, CONFIG, vcat(UNSUPPORTED_TESTS, LP_TESTS, BASIC))
 end
 
-# @testset "Integer Linear" begin
-#     MOIT.intlineartest(OPTIMIZER, CONFIG, [
-#         "indicator1", "indicator2", "indicator3", "indicator4", # indicator constraints not supported
-#         "semiconttest", "semiinttest", # semi integer vars not supported
-#         "int2" # SOS1 & SOS2 not supported
-#     ])
-# end
+const OPTIMIZER_CONSTRUCTOR = MOI.OptimizerWithAttributes(Coluna.Optimizer)#, MOI.Silent() => true) # MOI.Silent not supported
+const BRIDGED = MOI.instantiate(OPTIMIZER_CONSTRUCTOR, with_bridge_type = Float64)
+MOI.set(BRIDGED, MOI.RawParameter("default_optimizer"), GLPK.Optimizer)
+MOI.set(BRIDGED, MOI.RawParameter("params"), CL.Params(solver = ClA.SolveIpForm()))
+
+@testset "Integer Linear" begin
+    MOIT.intlineartest(BRIDGED, CONFIG, [
+        "indicator1", "indicator2", "indicator3", "indicator4", # indicator constraints not supported
+        "semiconttest", "semiinttest", # semi integer vars not supported
+        "int2" # SOS1 & SOS2 not supported
+    ])
+end
 
 # @testset "Unit LP" begin
 #     MOI.set(OPTIMIZER, MOI.RawParameter("params"), CL.Params(solver = ClA.SolveLpForm()))
@@ -180,5 +113,74 @@ end
 #     ])
 # end
 
-const OPTIMIZER_CONSTRUCTOR = MOI.OptimizerWithAttributes(Coluna.Optimizer)#, MOI.Silent() => true) # MOI.Silent not supported
-const BRIDGED = MOI.instantiate(OPTIMIZER_CONSTRUCTOR, with_bridge_type = Float64)
+@testset "Kpis" begin
+    data = CLD.GeneralizedAssignment.data("smallgap3.txt")
+
+    coluna = JuMP.optimizer_with_attributes(
+        CL.Optimizer,
+        "params" => CL.Params(
+            solver = ClA.TreeSearchAlgorithm(
+                conqueralg = ClA.ColCutGenConquer(
+                    colgen = ClA.ColumnGeneration(max_nb_iterations = 8)
+                ), maxnumnodes = 4
+            )
+        ),
+        "default_optimizer" => GLPK.Optimizer
+    )
+
+    problem, x, dec = CLD.GeneralizedAssignment.model(data, coluna)
+
+    JuMP.optimize!(problem)
+
+    @test MOI.get(problem, MOI.NodeCount()) == 4
+    @test isa(MOI.get(problem, MOI.SolveTime()), Float64)
+end
+
+@testset "branching_priority" begin
+    coluna = JuMP.optimizer_with_attributes(
+        Coluna.Optimizer,
+        "params" => CL.Params(solver=ClA.SolveIpForm()),
+        "default_optimizer" => GLPK.Optimizer
+    )
+    model = BlockModel(coluna, direct_model=true)
+    @variable(model, x)
+
+    @test BlockDecomposition.branchingpriority(model, x) == 1
+    BlockDecomposition.branchingpriority!(model, x, 2)
+    @test BlockDecomposition.branchingpriority(model, x) == 2
+
+    model2 = BlockModel(coluna)
+    @variable(model2, x)
+
+    @test BlockDecomposition.branchingpriority(model2, x) == 1
+    BlockDecomposition.branchingpriority!(model2, x, 2)
+    @test BlockDecomposition.branchingpriority(model2, x) == 2
+end
+
+@testset "write_to_file" begin
+    coluna = JuMP.optimizer_with_attributes(
+        Coluna.Optimizer,
+        "params" => CL.Params(solver=ClA.SolveIpForm()),
+        "default_optimizer" => GLPK.Optimizer
+    )
+    model = BlockModel(coluna, direct_model=true)
+    @variable(model, x)
+
+    @constraint(model, x <= 1)
+    @objective(model, Max, x)
+    optimize!(model)
+
+    dest = MOI.FileFormats.Model(format = MOI.FileFormats.FORMAT_MPS)
+    MOI.copy_to(dest, model)
+    MOI.write_to_file(dest, "model.mps")
+
+    filedata = MOI.FileFormats.Model(format = MOI.FileFormats.FORMAT_MPS)
+    MOI.read_from_file(filedata, "model.mps")
+
+    model2 = BlockModel(coluna, direct_model=true)
+    MOI.copy_to(model2, filedata)
+    optimize!(model2)
+
+    # the model is always written as a minimization problem
+    @test JuMP.objective_value(model) == -JuMP.objective_value(model2)
+end
diff --git a/test/user_algorithms_tests.jl b/test/user_algorithms_tests.jl
@@ -69,6 +69,8 @@ function Coluna.Algorithm.run!(
     heur_output = run!(algo.rm_heur, env, data, OptimizationInput(optstate))
     update_all_ip_primal_solutions!(optstate, getoptstate(heur_output))
 
+    setterminationstatus!(optstate, getterminationstatus(getoptstate(heur_output)))
+
     return OptimizationOutput(optstate)
 end