Implement permutation constraints in PyJobShop

Author: leonlan

pyjobshop/Model.py

@@ -38,6 +38,7 @@ def __init__(self):
         self._modes: list[Mode] = []
         self._constraints = Constraints()
         self._objective: Objective = Objective(weight_makespan=1)
+        self._permutation: bool = False
 
         self._id2job: dict[int, int] = {}
         self._id2resource: dict[int, int] = {}
@@ -190,6 +191,7 @@ def data(self) -> ProblemData:
             modes=self.modes,
             constraints=self.constraints,
             objective=self.objective,
+            permutation=self._permutation,
         )
 
     def add_job(
@@ -428,6 +430,12 @@ def set_objective(
         )
         return self._objective
 
+    def set_permutation(self, value: bool = True):
+        """
+        Sets the permutation property of the model.
+        """
+        self._permutation = value
+
     def solve(
         self,
         solver: str = "ortools",

pyjobshop/ProblemData.py

@@ -670,6 +670,9 @@ class ProblemData:
         constraints.
     objective
         The objective function. Default is minimizing the makespan.
+    permutation
+        Whether the problem is a permutation problem. Default is
+        ``False``.
     """
 
     def __init__(
@@ -680,6 +683,7 @@ def __init__(
         modes: list[Mode],
         constraints: Optional[Constraints] = None,
         objective: Optional[Objective] = None,
+        permutation: bool = False,
     ):
         self._jobs = jobs
         self._resources = resources
@@ -693,6 +697,7 @@ def __init__(
             if objective is not None
             else Objective(weight_makespan=1)
         )
+        self._permutation = permutation
 
         self._validate_parameters()
 
@@ -859,6 +864,13 @@ def objective(self) -> Objective:
         """
         return self._objective
 
+    @property
+    def permutation(self) -> bool:
+        """
+        Returns whether the problem is a permutation problem.
+        """
+        return self._permutation
+
     @property
     def num_jobs(self) -> int:
         """

pyjobshop/constants.py

@@ -1,3 +1,3 @@
-MAX_VALUE = 2**48
+MAX_VALUE = 2**32  # lowered for large-scale instances
 """int: Maximum allowed value, equal to 2^48.
 """

pyjobshop/solvers/cpoptimizer/Constraints.py

@@ -183,6 +183,25 @@ def _consecutive_constraints(self):
 
                     model.add(cpo.previous(seq_var, var1, var2))
 
+    def _same_sequence_constraints(self):
+        """
+        Adds same sequence constraints.
+        """
+        model, data = self._model, self._data
+
+        if not data.permutation:
+            return  # not a permutation problem
+
+        for idx in range(data.num_resources - 1):
+            seq_var1 = self._sequence_vars[idx]
+            seq_var2 = self._sequence_vars[idx + 1]
+
+            if seq_var1 is None or seq_var2 is None:
+                msg = "Requested resources do not have sequence variables."
+                raise ValueError(msg)
+
+            model.add(cpo.same_sequence(seq_var1, seq_var2))
+
     def add_constraints(self):
         """
         Adds all the constraints to the CP model.
@@ -195,3 +214,4 @@ def add_constraints(self):
         self._timing_constraints()
         self._identical_and_different_resource_constraints()
         self._consecutive_constraints()
+        self._same_sequence_constraints()

pyjobshop/solvers/ortools/Constraints.py

@@ -1,5 +1,6 @@
 import numpy as np
 from ortools.sat.python.cp_model import CpModel, LinearExpr
+from itertools import product
 
 import pyjobshop.solvers.utils as utils
 from pyjobshop.ProblemData import (
@@ -203,48 +204,54 @@ def _circuit_constraints(self):
         """
         Creates the circuit constraints for each machine, if activated by
         sequencing constraints (consecutive and setup times).
+
+        IMPORTANT: This is specifically implemented for the experiments in the
+        paper and it is not meant to be used outside the scope of those
+        experiments because it may not be compatible with all other features.
         """
         model, data = self._model, self._data
         setup_times = utils.setup_times_matrix(data)
 
-        for idx, resource in enumerate(data.resources):
-            if not isinstance(resource, Machine):
-                continue
+        if not data.permutation:
+            return  # not a permutation problem, skip
 
-            seq_var = self._sequence_vars[idx]
-            if not seq_var.is_active:
-                # No sequencing constraints active. Skip the creation of
-                # expensive circuit constraints.
-                continue
+        # Create arcs for circuit constraints.
+        arcs = []
+        for idx1 in range(data.num_jobs):
+            arcs.append((0, idx1 + 1, model.new_bool_var("start")))
+            arcs.append((idx1 + 1, 0, model.new_bool_var("end")))
 
-            mode_vars = seq_var.mode_vars
-            arcs = seq_var.arcs
+        lits = {}
+        for idx1, idx2 in product(range(data.num_jobs), repeat=2):
+            if idx1 != idx2:
+                lit = model.new_bool_var(f"{idx1} -> {idx2}")
+                lits[idx1, idx2] = lit
+                arcs.append((idx1 + 1, idx2 + 1, lit))
 
-            graph = [(u, v, var) for (u, v), var in arcs.items()]
-            model.add_circuit(graph)
+        model.add_circuit(arcs)
 
-            for idx1, var1 in enumerate(mode_vars):
-                # If the (dummy) self arc is selected, then the var must not
-                # be present.
-                model.add(arcs[idx1, idx1] <= ~var1.present)
-                model.add(arcs[seq_var.DUMMY, seq_var.DUMMY] <= ~var1.present)
+        for res_idx, resource in enumerate(data.resources):
+            if not isinstance(resource, Machine):
+                raise ValueError("Machines only in permutation problems.")
 
-            for idx1, var1 in enumerate(mode_vars):
-                for idx2, var2 in enumerate(mode_vars):
-                    if idx1 == idx2:
-                        continue
+            seq_var = self._sequence_vars[res_idx]
+            assert seq_var is not None
+
+            for idx1, idx2 in product(range(data.num_jobs), repeat=2):
+                if idx1 == idx2:
+                    continue
 
-                    # If the arc is selected, then both tasks must be present.
-                    model.add(arcs[idx1, idx2] <= var1.present)
-                    model.add(arcs[idx1, idx2] <= var2.present)
-
-                    setup = (
-                        setup_times[idx, var1.task_idx, var2.task_idx]
-                        if setup_times is not None
-                        else 0
-                    )
-                    expr = var1.end + setup <= var2.start
-                    model.add(expr).only_enforce_if(arcs[idx1, idx2])
+                var1 = seq_var.mode_vars[idx1]
+                var2 = seq_var.mode_vars[idx2]
+
+                lit = lits[idx1, idx2]
+                setup = (
+                    setup_times[res_idx, var1.task_idx, var2.task_idx]
+                    if setup_times is not None
+                    else 0
+                )
+                expr = var1.end + setup <= var2.start
+                model.add(expr).only_enforce_if(lit)
 
     def add_constraints(self):
         """