Add status to lp solvers (and default one to all solvers)

Gurobi and mathopt give information about solver status. As for minizinc
and cpsat, we make it available on the wrapper level.

As we use the same enumeration `StatusSolver`, we move it to `solver_do`
module. And thus add for all solvers a `status_solver` attribute, by
default equal to unknown and which is used by `is_optimal()` method.

We remove `get_status_solver()` method which was only returning
`self.status_solver`.

discrete_optimization/generic_tools/cp_tools.py

@@ -23,7 +23,7 @@
     CallbackList,
 )
 from discrete_optimization.generic_tools.do_problem import Solution
-from discrete_optimization.generic_tools.do_solver import SolverDO
+from discrete_optimization.generic_tools.do_solver import SolverDO, StatusSolver
 from discrete_optimization.generic_tools.exceptions import SolveEarlyStop
 from discrete_optimization.generic_tools.hyperparameters.hyperparameter import (
     EnumHyperparameter,
@@ -160,13 +160,6 @@ class SignEnum(Enum):
     UP = ">"
 
 
-class StatusSolver(Enum):
-    SATISFIED = "SATISFIED"
-    UNSATISFIABLE = "UNSATISFIABLE"
-    OPTIMAL = "OPTIMAL"
-    UNKNOWN = "UNKNOWN"
-
-
 map_mzn_status_to_do_status: dict[Status, StatusSolver] = {
     Status.SATISFIED: StatusSolver.SATISFIED,
     Status.UNSATISFIABLE: StatusSolver.UNSATISFIABLE,
@@ -180,22 +173,6 @@ class CPSolver(SolverDO):
     Additional function to be implemented by a CP Solver.
     """
 
-    status_solver: Optional[StatusSolver] = None
-
-    def is_optimal(self) -> Optional[bool]:
-        """Tell if found solution is supposed to be optimal.
-
-        To be called after a solve.
-
-        Returns:
-            optimality of the solution. If information missing, returns None instead.
-
-        """
-        if self.status_solver is None or self.status_solver == StatusSolver.UNKNOWN:
-            return None
-        else:
-            return self.status_solver == StatusSolver.OPTIMAL
-
     @abstractmethod
     def init_model(self, **args: Any) -> None:
         """
@@ -215,9 +192,6 @@ def solve(
     ) -> ResultStorage:
         ...
 
-    def get_status_solver(self) -> Union[StatusSolver, None]:
-        return self.status_solver
-
 
 class MinizincCPSolver(CPSolver):
     """CP solver wrapping a minizinc solver."""

discrete_optimization/generic_tools/do_solver.py

@@ -7,6 +7,7 @@
 
 from abc import ABC, abstractmethod
 from collections.abc import Iterable
+from enum import Enum
 from typing import Any, Optional
 
 from discrete_optimization.generic_tools.callbacks.callback import Callback
@@ -28,10 +29,18 @@
 )
 
 
+class StatusSolver(Enum):
+    SATISFIED = "SATISFIED"
+    UNSATISFIABLE = "UNSATISFIABLE"
+    OPTIMAL = "OPTIMAL"
+    UNKNOWN = "UNKNOWN"
+
+
 class SolverDO(Hyperparametrizable, ABC):
     """Base class for a discrete-optimization solver."""
 
     problem: Problem
+    status_solver: StatusSolver = StatusSolver.UNKNOWN
 
     def __init__(
         self,
@@ -102,7 +111,10 @@ def is_optimal(self) -> Optional[bool]:
             optimality of the solution. If information missing, returns None instead.
 
         """
-        return None
+        if self.status_solver == StatusSolver.UNKNOWN:
+            return None
+        else:
+            return self.status_solver == StatusSolver.OPTIMAL
 
     def get_model_objectives_available(self) -> list[str]:
         """List objectives available for lexico optimization

discrete_optimization/generic_tools/lp_tools.py

@@ -7,6 +7,7 @@
 import datetime
 import logging
 from abc import abstractmethod
+from collections import defaultdict
 from collections.abc import Callable
 from enum import Enum
 from typing import Any, Optional, Union
@@ -23,7 +24,11 @@
     Problem,
     Solution,
 )
-from discrete_optimization.generic_tools.do_solver import SolverDO, WarmstartMixin
+from discrete_optimization.generic_tools.do_solver import (
+    SolverDO,
+    StatusSolver,
+    WarmstartMixin,
+)
 from discrete_optimization.generic_tools.exceptions import SolveEarlyStop
 from discrete_optimization.generic_tools.hyperparameters.hyperparameter import (
     EnumHyperparameter,
@@ -41,7 +46,14 @@
     gurobi_available = False
 else:
     gurobi_available = True
-    GRB = gurobipy.GRB
+    map_gurobi_status_to_do_status: dict[int, StatusSolver] = defaultdict(
+        lambda: StatusSolver.UNKNOWN,
+        {
+            gurobipy.GRB.status.OPTIMAL: StatusSolver.OPTIMAL,
+            gurobipy.GRB.status.INFEASIBLE: StatusSolver.UNSATISFIABLE,
+            gurobipy.GRB.status.SUBOPTIMAL: StatusSolver.SATISFIED,
+        },
+    )
 
 try:
     import docplex
@@ -511,6 +523,8 @@ def optimize_model(
             cb=mathopt_cb,
         )
         self.termination = mathopt_res.termination
+        self.status_solver = map_mathopt_status_to_do_status[self.termination.reason]
+
         logger.info(f"Solver found {len(mathopt_res.solutions)} solutions")
         if mathopt_res.termination.reason in [
             mathopt.TerminationReason.OPTIMAL,
@@ -520,6 +534,19 @@ def optimize_model(
         return mathopt_res
 
 
+map_mathopt_status_to_do_status: dict[mathopt.TerminationReason, StatusSolver] = {
+    mathopt.TerminationReason.OPTIMAL: StatusSolver.OPTIMAL,
+    mathopt.TerminationReason.INFEASIBLE: StatusSolver.UNSATISFIABLE,
+    mathopt.TerminationReason.INFEASIBLE_OR_UNBOUNDED: StatusSolver.UNKNOWN,
+    mathopt.TerminationReason.UNBOUNDED: StatusSolver.UNKNOWN,
+    mathopt.TerminationReason.FEASIBLE: StatusSolver.SATISFIED,
+    mathopt.TerminationReason.NO_SOLUTION_FOUND: StatusSolver.UNSATISFIABLE,
+    mathopt.TerminationReason.IMPRECISE: StatusSolver.UNKNOWN,
+    mathopt.TerminationReason.NUMERICAL_ERROR: StatusSolver.UNKNOWN,
+    mathopt.TerminationReason.OTHER_ERROR: StatusSolver.UNKNOWN,
+}
+
+
 def _mathopt_cb_get_obj_value_for_current_solution():
     raise RuntimeError("Cannot retrieve objective!")
 
@@ -652,6 +679,7 @@ def optimize_model(
             parameters_milp=parameters_milp, time_limit=time_limit, **kwargs
         )
         self.model.optimize()
+        self.status_solver = map_gurobi_status_to_do_status[self.model.Status]
 
         logger.info(f"Problem has {self.model.NumObj} objectives")
         logger.info(f"Solver found {self.model.SolCount} solutions")
@@ -724,13 +752,13 @@ def __init__(self, do_solver: GurobiMilpSolver, callback: Callback):
         self.nb_solutions = 0
 
     def __call__(self, model, where) -> None:
-        if where == GRB.Callback.MIPSOL:
+        if where == gurobipy.GRB.Callback.MIPSOL:
             try:
                 # retrieve and store new solution
                 sol = self.do_solver.retrieve_current_solution(
                     get_var_value_for_current_solution=model.cbGetSolution,
                     get_obj_value_for_current_solution=lambda: model.cbGet(
-                        GRB.Callback.MIPSOL_OBJ
+                        gurobipy.GRB.Callback.MIPSOL_OBJ
                     ),
                 )
                 fit = self.do_solver.aggreg_from_sol(sol)

discrete_optimization/generic_tools/ortools_cpsat_tools.py

@@ -21,12 +21,9 @@
     Callback,
     CallbackList,
 )
-from discrete_optimization.generic_tools.cp_tools import (
-    CPSolver,
-    ParametersCP,
-    StatusSolver,
-)
+from discrete_optimization.generic_tools.cp_tools import CPSolver, ParametersCP
 from discrete_optimization.generic_tools.do_problem import Solution
+from discrete_optimization.generic_tools.do_solver import StatusSolver
 from discrete_optimization.generic_tools.exceptions import SolveEarlyStop
 from discrete_optimization.generic_tools.result_storage.result_storage import (
     ResultStorage,

discrete_optimization/rcpsp/solver/cpsat_solver.py

@@ -19,9 +19,8 @@
     ObjLinearExprT,
 )
 
-from discrete_optimization.generic_tools.cp_tools import StatusSolver
 from discrete_optimization.generic_tools.do_problem import ParamsObjectiveFunction
-from discrete_optimization.generic_tools.do_solver import WarmstartMixin
+from discrete_optimization.generic_tools.do_solver import StatusSolver, WarmstartMixin
 from discrete_optimization.generic_tools.hyperparameters.hyperparameter import (
     CategoricalHyperparameter,
 )

examples/coloring/coloring_cpspat_solver_example.py

@@ -47,7 +47,7 @@ def run_cpsat_coloring():
         callbacks=[NbIterationTracker(step_verbosity_level=logging.INFO)],
         parameters_cp=p,
     )
-    print("Status solver : ", solver.get_status_solver())
+    print("Status solver : ", solver.status_solver)
     solution, fit = result_store.get_best_solution_fit()
     plot_coloring_solution(solution)
     plt.show()
@@ -70,7 +70,7 @@ def run_cpsat_coloring_with_constraints():
     p = ParametersCP.default()
     result_store = solver.solve(parameters_cp=p, time_limit=20)
     solution, fit = result_store.get_best_solution_fit()
-    print("Status solver : ", solver.get_status_solver())
+    print("Status solver : ", solver.status_solver)
     plot_coloring_solution(solution)
     plt.show()
     print(solution, fit)

examples/jsp/jsp_cpsat_example.py

@@ -31,7 +31,7 @@ def run_cpsat_jsp():
     p.nb_process = 10
     res = solver.solve(parameters_cp=p, time_limit=10)
     sol = res.get_best_solution_fit()[0]
-    print(solver.get_status_solver())
+    print(solver.status_solver)
     assert problem.satisfy(sol)
     print(problem.evaluate(sol))
 

examples/knapsack/knapsack_cpsat_solver.py

@@ -20,7 +20,7 @@ def run_cpsat_knapsack():
     params_cp = ParametersCP.default()
     result_storage = cp_model.solve(parameters_cp=params_cp, time_limit=10)
     sol, fit = result_storage.get_best_solution_fit()
-    print("Status solver :", cp_model.get_status_solver())
+    print("Status solver :", cp_model.status_solver)
     assert knapsack_model.satisfy(sol)
 
 

examples/maximum_independent_set/lns_cpsat_mis.py

@@ -50,7 +50,7 @@ def run_lns():
     )
     sol, fit = res.get_best_solution_fit()
     print("Found by cpsat :", fit)
-    print("Status : ", solver.get_status_solver())
+    print("Status : ", solver.status_solver)
 
     initial_solution_provider = TrivialInitialSolution(
         solver.create_result_storage(
@@ -106,7 +106,7 @@ def run_lns_mix():
     )
     sol, fit = res.get_best_solution_fit()
     print("Found by cpsat :", fit)
-    print("Status : ", solver.get_status_solver())
+    print("Status : ", solver.status_solver)
 
     initial_solution_provider = TrivialInitialSolution(
         solver.create_result_storage(

examples/rcpsp_multiskill/mslib/mslib_parse_and_solve.py

@@ -60,7 +60,7 @@ def example_mslib_cpsat():
         ],
         ortools_cpsat_solver_kwargs={"log_search_progress": True},
     )
-    print(solver.get_status_solver())
+    print(solver.status_solver)
     from discrete_optimization.rcpsp_multiskill.plots.plot_solution import (
         plot_resource_individual_gantt,
     )

examples/vrp/vrp_cpsat_example.py

@@ -26,7 +26,7 @@ def run_cpsat_vrp():
     p = ParametersCP.default_cpsat()
     p.nb_process = 10
     res = solver.solve(parameters_cp=p, time_limit=20)
-    print(solver.get_status_solver())
+    print(solver.status_solver)
     sol, fit = res.get_best_solution_fit()
     sol: VrpSolution
     print(problem.evaluate(sol))
@@ -111,7 +111,7 @@ def warm_starting():
             fix_variables_to_their_hinted_value=False, log_search_progress=True
         ),
     )
-    print(solver.get_status_solver())
+    print(solver.status_solver)
     # assert res[0][0].list_paths == start_solution.list_paths
 
 

tests/coloring/test_coloring_lns_cpsat.py

@@ -76,7 +76,7 @@ def test_lns_cpsat_coloring():
         parameters_cp=p,
         time_limit_subsolver=20,
     )
-    print("Status solver : ", solver.get_status_solver())
+    print("Status solver : ", solver.status_solver)
     solution, fit = result_store.get_best_solution_fit()
     # plot_coloring_solution(solution)
     # plt.show()
@@ -128,7 +128,7 @@ def test_lns_cpsat_coloring_with_constraints():
         ],
         parameters_cp=p,
     )
-    print("Status solver : ", solver.get_status_solver())
+    print("Status solver : ", solver.status_solver)
     solution, fit = result_store.get_best_solution_fit()
     # plot_coloring_solution(solution)
     # plt.show()