New solvers based on dynamic programming and the didppy library #321

g-poveda · 2024-10-10T14:28:18Z

create a generic solver class with Did library aligned with our existing one
create base model for jsp, fjsp, facility, coloring, mis, rcpsp, tsp, vrp

nhuet

I think that some changes should be put in another PR (it seems they do not have a relation with dynamic programming).
It would be nice to test the callback use at least in one of the did solvers to check this is working as intended (logging, stopping, crashing if callback raise an error).

nhuet · 2024-10-11T07:41:14Z

discrete_optimization/fjsp/solvers/__init__.py

@@ -0,0 +1,3 @@
+#  Copyright (c) 2024 AIRBUS and its affiliates.
+#  This source code is licensed under the MIT license found in the
+#  LICENSE file in the root directory of this source tree.


Why is this in the PR ?

nhuet · 2024-10-11T07:41:23Z

discrete_optimization/fjsp/solvers/cpsat_lns_fjsp_solver.py

@@ -79,6 +83,7 @@ def adding_constraint_from_results_store(
            # constraints.append(solver.cp_model.Add(solver.variables["ends"][k] <= end+20))
            # constraints.append(solver.cp_model.Add(solver.variables["ends"][k] >= end-20))
        solver.cp_model.Minimize(solver.variables["makespan"])
+        solver.set_warm_start(sol)


Why is this in the PR ?

nhuet · 2024-10-11T07:41:30Z

discrete_optimization/fjsp/solvers/cpsat_lns_fjsp_solver.py

@@ -114,4 +119,5 @@ def adding_constraint_from_results_store(
        # m = solver.cp_model.NewIntVar(lb=0, ub=makespan, name="")
        # solver.cp_model.AddMaxEquality(m, [solver.variables["ends"][k] for k in keys_part])
        # solver.cp_model.Minimize(m) #+sum([solver.variables["ends"][k] for k in keys_part]))
+        solver.set_warm_start(sol)


Why is this in the PR ?

nhuet · 2024-10-11T07:41:43Z

discrete_optimization/facility/facility_model.py

@@ -148,7 +148,7 @@ def evaluate(self, variable: FacilitySolution) -> dict[str, float]:  # type: ign
        d = self.evaluate_cost(variable)
        capacity_constraint_violation = 0
        for f in d["details"]:
-            capacity_constraint_violation = max(
+            capacity_constraint_violation += max(


Why is this in the PR ?

nhuet · 2024-10-11T07:42:19Z

discrete_optimization/vrp/solver/solver_ortools.py

@@ -56,7 +56,7 @@ class LocalSearchMetaheuristic(Enum):
 class BasicRoutingMonitor:
    def __init__(self, do_solver: "VrpORToolsSolver"):
        self.model = do_solver.routing
-        self.problem = do_solver.vrp_model
+        self.problem = do_solver.problem


Why is this in the PR ?

nhuet · 2024-10-11T07:42:24Z

discrete_optimization/vrp/solver/solver_ortools.py

@@ -169,7 +169,7 @@ def retrieve(self, solution: pywrapcp.Assignment) -> VrpSolution:
            vehicle_tours_all[-1] += [self.manager.IndexToNode(index)]
        logger.debug(f"Route distance : {route_distance}")
        logger.debug(f"Vehicle tours : {vehicle_tours}")
-        logger.debug(f"Objective : {objective}")
+        logger.info(f"Objective : {objective}")


Why is this in the PR ?

nhuet · 2024-10-11T07:42:35Z

examples/coloring/coloring_asp_solver.py

+                step_verbosity_level=logging.INFO, end_verbosity_level=logging.INFO
+            )
+        ],
+        time_limit=5,


Why is this in the PR ?

nhuet · 2024-10-11T07:42:42Z

examples/coloring/coloring_cpspat_solver_example.py

@@ -39,7 +42,7 @@ def run_cpsat_coloring():
    p = ParametersCP.default_cpsat()
    logging.info("Starting solve")
    result_store = solver.solve(
-        callbacks=[NbIterationTracker(step_verbosity_level=logging.INFO)],
+        callbacks=[ObjectiveLogger(step_verbosity_level=logging.INFO)],


Why is this in the PR ?

nhuet · 2024-10-11T07:42:54Z

examples/knapsack/knapsack_cpsat_solver.py

@@ -13,7 +13,7 @@


 def run_cpsat_knapsack():
-    file = [f for f in get_data_available() if "ks_1000_0" in f][0]
+    file = [f for f in get_data_available() if "ks_10000_0" in f][0]


Why is this in the PR ?

nhuet · 2024-10-11T07:43:32Z

examples/vrp/vrp_solver_example.py

@@ -26,7 +30,11 @@ def run_ortools_vrp_solver():
        first_solution_strategy=FirstSolutionStrategy.SAVINGS,
        local_search_metaheuristic=LocalSearchMetaheuristic.GUIDED_LOCAL_SEARCH,
    )
-    res = solver.solve(time_limit_seconds=20)
+    res = solver.solve(


Why is this in the PR ?

-adding didppy models available for many DO problems -implemented examples, tests for each implemented problem

nhuet reviewed Oct 11, 2024

View reviewed changes

g-poveda force-pushed the gpd/ddpy_dev branch from aea5bc7 to a53f28c Compare October 11, 2024 12:24

Dynamic programming solver implementation

e4e676f

-adding didppy models available for many DO problems -implemented examples, tests for each implemented problem

g-poveda force-pushed the gpd/ddpy_dev branch from a53f28c to e4e676f Compare October 11, 2024 12:51

g-poveda merged commit 24fa8e7 into airbus:master Oct 11, 2024
5 of 12 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

New solvers based on dynamic programming and the didppy library #321

New solvers based on dynamic programming and the didppy library #321

g-poveda commented Oct 10, 2024

nhuet left a comment

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

nhuet Oct 11, 2024

New solvers based on dynamic programming and the didppy library #321

New solvers based on dynamic programming and the didppy library #321

Conversation

g-poveda commented Oct 10, 2024

nhuet left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment