Make Julia's solution format consistent with Python's

src/julia/src/MIPLearn.jl

@@ -15,36 +15,49 @@ Instance = miplearn.Instance
 LearningSolver = miplearn.LearningSolver
 InternalSolver = miplearn.solvers.internal.InternalSolver
 
+function varname_split(varname::String)
+    m = match(r"([^[]*)\[(.*)\]", varname)
+    return m.captures[1], m.captures[2]
+end
+
 @pydef mutable struct JuMPSolver <: InternalSolver
     function __init__(self; optimizer=CPLEX.Optimizer)
         self.optimizer = optimizer
     end
 
     function add_constraint(self, constraint)
-        @error "Not implemented"
+        @error "JuMPSolver: add_constraint not implemented"
     end
 
     function set_warm_start(self, solution)
-        for (varname, value) in solution
+        for (basename, subsolution) in solution
-            var = JuMP.variable_by_name(self.model, varname)
+            for (idx, value) in subsolution
-            JuMP.set_start_value(var, value)
+                var = self.basename_idx_to_var[basename, idx]
+                JuMP.set_start_value(var, value)
+            end
         end
     end
 
     function clear_warm_start(self)
-        @error "Not implemented"
+        @error "JuMPSolver: clear_warm_start not implemented"
     end
 
     function fix(self, solution)
-        for (varname, value) in solution
+        for (basename, subsolution) in solution
-            var = JuMP.variable_by_name(self.model, varname)
+            for (idx, value) in subsolution
-            JuMP.fix(var, value, force=true)
+                var = self.basename_idx_to_var[basename, idx]
+                JuMP.fix(var, value, force=true)
+            end
         end
     end
 
     function set_instance(self, instance, model)
         self.instance = instance
         self.model = model
+        self.var_to_basename_idx = Dict(var => varname_split(JuMP.name(var))
+                                        for var in JuMP.all_variables(self.model))
+        self.basename_idx_to_var = Dict(varname_split(JuMP.name(var)) => var
+                                        for var in JuMP.all_variables(self.model))
         self.bin_vars = [var
                          for var in JuMP.all_variables(self.model)
                          if JuMP.is_binary(var)]
@@ -102,20 +115,27 @@ InternalSolver = miplearn.solvers.internal.InternalSolver
     end
 
     function _update_solution(self)
-        self.solution =  Dict(JuMP.name(var) => JuMP.value(var)
+        solution = Dict()
-                              for var in JuMP.all_variables(self.model))
+        for var in JuMP.all_variables(self.model)
+            basename, idx = self.var_to_basename_idx[var]
+            if !haskey(solution, basename)
+                solution[basename] = Dict()
+            end
+            solution[basename][idx] = JuMP.value(var)
+        end
+        self.solution = solution
     end
 
     function set_gap_tolerance(self, gap_tolerance)
-        @error "Not implemented"
+        @error "JuMPSolver: set_gap_tolerance not implemented"
     end
 
     function set_node_limit(self)
-        @error "Not implemented"
+        @error "JuMPSolver: set_node_limit not implemented"
     end
 
     function set_threads(self, threads)
-        @error "Not implemented"
+        @error "JuMPSolver: set_threads not implemented"
     end
 
     function set_time_limit(self, time_limit)
@@ -148,6 +168,6 @@ end
     end
 end
 
-export JuMPSolver, KnapsackInstance
+export LearningSolver, JuMPSolver, KnapsackInstance
 
 end # module

src/julia/test/jump_solver.jl

@@ -7,6 +7,11 @@ using MIPLearn
 using CPLEX
 using Gurobi
 
+@testset "varname_split" begin
+    @test MIPLearn.varname_split("x[1]") == ("x", "1")
+end
+
+
 @testset "JuMPSolver" begin
     for optimizer in [CPLEX.Optimizer, Gurobi.Optimizer]
         instance = KnapsackInstance([23., 26., 20., 18.],
@@ -17,12 +22,12 @@ using Gurobi
         solver = JuMPSolver(optimizer=optimizer)
         solver.set_instance(instance, model)
         solver.set_time_limit(30)
-        solver.set_warm_start(Dict(
+        solver.set_warm_start(Dict("x" => Dict(
-            "x[1]" => 1.0,
+            "1" => 1.0,
-            "x[2]" => 0.0,
+            "2" => 0.0,
-            "x[3]" => 0.0,
+            "3" => 0.0,
-            "x[4]" => 1.0,
+            "4" => 1.0,
-        ))
+        )))
         stats = solver.solve()
 
         @test stats["Lower bound"] == 1183.0
@@ -31,26 +36,26 @@ using Gurobi
         @test stats["Wallclock time"] > 0
 
         solution = solver.get_solution()
-        @test solution["x[1]"] == 1.0
+        @test solution["x"]["1"] == 1.0
-        @test solution["x[2]"] == 0.0
+        @test solution["x"]["2"] == 0.0
-        @test solution["x[3]"] == 1.0
+        @test solution["x"]["3"] == 1.0
-        @test solution["x[4]"] == 1.0
+        @test solution["x"]["4"] == 1.0
 
         stats = solver.solve_lp()
         @test round(stats["Optimal value"], digits=3) == 1287.923
 
         solution = solver.get_solution()
-        @test round(solution["x[1]"], digits=3) == 1.000
+        @test round(solution["x"]["1"], digits=3) == 1.000
-        @test round(solution["x[2]"], digits=3) == 0.923
+        @test round(solution["x"]["2"], digits=3) == 0.923
-        @test round(solution["x[3]"], digits=3) == 1.000
+        @test round(solution["x"]["3"], digits=3) == 1.000
-        @test round(solution["x[4]"], digits=3) == 0.000
+        @test round(solution["x"]["4"], digits=3) == 0.000
 
-        solver.fix(Dict(
+        solver.fix(Dict("x" => Dict(
-            "x[1]" => 1.0,
+            "1" => 1.0,
-            "x[2]" => 0.0,
+            "2" => 0.0,
-            "x[3]" => 0.0,
+            "3" => 0.0,
-            "x[4]" => 1.0,
+            "4" => 1.0,
-        ))
+        )))
         stats = solver.solve()
         @test stats["Lower bound"] == 725.0
         @test stats["Upper bound"] == 725.0

src/julia/test/learning_solver.jl

@@ -0,0 +1,50 @@
+#  MIPLearn: Extensible Framework for Learning-Enhanced Mixed-Integer Optimization
+#  Copyright (C) 2020, UChicago Argonne, LLC. All rights reserved.
+#  Released under the modified BSD license. See COPYING.md for more details.
+
+using Test
+using MIPLearn
+using CPLEX
+using Gurobi
+
+@testset "LearningSolver" begin
+    for optimizer in [CPLEX.Optimizer, Gurobi.Optimizer]
+        instance = KnapsackInstance([23., 26., 20., 18.],
+                                    [505., 352., 458., 220.],
+                                    67.0)
+        model = instance.to_model()
+
+        solver = LearningSolver(solver=JuMPSolver(optimizer=optimizer))
+        stats = solver.solve(instance, model)
+
+        @test stats["Lower bound"] == 1183.0
+        @test stats["Upper bound"] == 1183.0
+        @test stats["Sense"] == "max"
+        @test stats["Wallclock time"] > 0
+
+#         solution = solver.get_solution()
+#         @test solution["x[1]"] == 1.0
+#         @test solution["x[2]"] == 0.0
+#         @test solution["x[3]"] == 1.0
+#         @test solution["x[4]"] == 1.0
+#
+#         stats = solver.solve_lp()
+#         @test round(stats["Optimal value"], digits=3) == 1287.923
+#
+#         solution = solver.get_solution()
+#         @test round(solution["x[1]"], digits=3) == 1.000
+#         @test round(solution["x[2]"], digits=3) == 0.923
+#         @test round(solution["x[3]"], digits=3) == 1.000
+#         @test round(solution["x[4]"], digits=3) == 0.000
+#
+#         solver.fix(Dict(
+#             "x[1]" => 1.0,
+#             "x[2]" => 0.0,
+#             "x[3]" => 0.0,
+#             "x[4]" => 1.0,
+#         ))
+#         stats = solver.solve()
+#         @test stats["Lower bound"] == 725.0
+#         @test stats["Upper bound"] == 725.0
+    end
+end

src/julia/test/runtests.jl

@@ -3,10 +3,8 @@
 #  Released under the modified BSD license. See COPYING.md for more details.
 
 using Test
-using MIPLearn
-using CPLEX
-using Gurobi
 
 @testset "MIPLearn" begin
     include("jump_solver.jl")
+    #include("learning_solver.jl")
 end

src/python/miplearn/solvers/learning.py

@@ -79,9 +79,8 @@ class LearningSolver:
             solver = GurobiSolver()
         elif callable(self.internal_solver_factory):
             solver = self.internal_solver_factory()
-            assert isinstance(solver, InternalSolver)
         else:
-            raise Exception("solver %s not supported" % self.internal_solver_factory)
+            solver = self.internal_solver_factory
         solver.set_threads(self.threads)
         if self.time_limit is not None:
             solver.set_time_limit(self.time_limit)