scip interface (first version)

.github/workflows/python-test.yml

@@ -19,6 +19,7 @@ jobs:
               sudo snap install minizinc --classic
           - solver: pindakaas
           - solver: pumpkin
+          - solver: scip
     uses: ./.github/workflows/test-solver.yml
     with:
       solver: ${{ matrix.solver }}

README.md

@@ -40,7 +40,7 @@ Install simply with `pip install cpmpy`
 CPMpy can translate to a wide variety of constraint solving paradigms, including both commercial and open-source solvers.
 
 * **CP Solvers**: OR-Tools (default), IBM CP Optimizer (license required), Choco, Glasgow GCS, Pumpkin, MiniZinc+solvers
-* **ILP Solvers**: Gurobi (license required), CPLEX (license required)
+* **ILP Solvers**: SCIP, Gurobi (license required), CPLEX (license required)
 * **GO Solvers**: Hexaly (license required)
 * **SMT Solvers**: Z3
 * **PB Solvers**: Exact

cpmpy/solvers/__init__.py

@@ -43,6 +43,7 @@
     cplex
     hexaly
     rc2
+    scip
 
 =========================
 List of helper submodules
@@ -78,6 +79,7 @@
 from .cplex import CPM_cplex
 from .hexaly import CPM_hexaly
 from .rc2 import CPM_rc2
+from .scip import CPM_scip
 
 __all__ = [
     "CPM_choco",
@@ -94,6 +96,7 @@
     "CPM_pysat",
     "CPM_pysdd",
     "CPM_rc2",
+    "CPM_scip",
     "CPM_z3",
     "SolverLookup",
     "param_combinations",

cpmpy/solvers/utils.py

@@ -22,6 +22,7 @@
 from .ortools import CPM_ortools
 from .minizinc import CPM_minizinc
 from .pysat import CPM_pysat
+from .scip import CPM_scip
 from .z3 import CPM_z3
 from .gcs import CPM_gcs
 from .pysdd import CPM_pysdd
@@ -90,6 +91,7 @@ def base_solvers(cls):
                 ("pindakaas", CPM_pindakaas),
                 ("hexaly", CPM_hexaly),
                 ("rc2", CPM_rc2),
+                ("scip", CPM_scip),
                ]
 
     @classmethod

docs/installation_instructions.rst

@@ -43,7 +43,7 @@ CPMpy supports a multitude of solvers of different technologies to be used as ba
 .. code-block:: bash
 
     # Choose any subset of solvers to install
-    $ pip install cpmpy[choco, cpo, exact, gcs, gurobi, minizinc, pysat, pysdd, z3] 
+    $ pip install cpmpy[choco, cpo, exact, gcs, gurobi, minizinc, pysat, pysdd, z3, scip] 
 
 Some solvers require additional steps (like acquiring a (aca.) license). Have a look at :ref:`this <supported-solvers>` overview.
 

docs/solvers.md

@@ -2,7 +2,7 @@
 
 CPMpy can be used as a declarative modeling language: you create a `Model()`, add constraints and call `solve()` on it.
 
-The default solver is OR-Tools CP-SAT, an award winning constraint solver. But CPMpy supports multiple other solvers: more CP solvers (IBM-CPO, Choco, GCS, Pumpkin), ILP solvers (Gurobi, IBM ILOG CPLEX), SAT solvers (those in PySAT and Pindakaas), the RC2 MaxSAT solver, the Z3 SMT solver, a conflict-driven cutting-planes solver (Exact), even a knowledge compiler (PySDD), a global optimisation solver (Hexaly) and any CP solver supported by the text-based MiniZinc language.
+The default solver is OR-Tools CP-SAT, an award winning constraint solver. But CPMpy supports multiple other solvers: more CP solvers (IBM-CPO, Choco, GCS, Pumpkin), ILP solvers (Gurobi, IBM ILOG CPLEX, SCIP), SAT solvers (those in PySAT and Pindakaas), the RC2 MaxSAT solver, the Z3 SMT solver, a conflict-driven cutting-planes solver (Exact), even a knowledge compiler (PySDD), a global optimisation solver (Hexaly) and any CP solver supported by the text-based MiniZinc language.
 
 See the list of solvers known by CPMpy with:
 

mypy.ini

@@ -58,6 +58,9 @@ ignore_missing_imports = True
 [mypy-z3.*]
 ignore_missing_imports = True
 
+[mypy-pyscipopt.*]
+ignore_missing_imports = True
+
 # untyped pycsp3, used in tools.xcsp3
 [mypy-pycsp3.*]
-ignore_missing_imports = True
+ignore_missing_imports = True

setup.py

@@ -33,6 +33,7 @@ def get_version(rel_path):
     "pumpkin": ["pumpkin-solver>=0.3.0"], # CPMpy requires features only available from Pumpkin version >=0.3.0
     "pindakaas": ["pindakaas>=0.5.0"],
     "cplex": ["docplex>=2.28.240", "cplex>=20.1.0.4"],
+    "scip": ["pyscipopt>=6.1"]
 }
 solver_dependencies["all"] = list({pkg for group in solver_dependencies.values() for pkg in group}) 
 

tests/test_solvers.py

@@ -18,6 +18,7 @@
 from cpmpy.solvers.exact import CPM_exact
 from cpmpy.solvers.choco import CPM_choco
 from cpmpy.solvers.cplex import CPM_cplex
+from cpmpy.solvers.scip import CPM_scip
 from cpmpy import SolverLookup
 from cpmpy.exceptions import MinizincNameException, NotSupportedError
 
@@ -1008,7 +1009,6 @@ def test_incremental_assumptions(self, solver):
         assert s.solve(assumptions=[])
 
     def test_vars_not_removed(self, solver):
-
         bvs = cp.boolvar(shape=3)
         m = cp.Model([cp.any(bvs) <= 2])
 
@@ -1050,13 +1050,8 @@ def test_partial_div_mod(self, solver):
         m += x // y == d
         m += x % y == r
         sols = set()
-        solution_limit = None
-        time_limit = None
-        if solver == 'gurobi':
-            solution_limit = 15 # Gurobi does not like this model, and gets stuck finding all solutions
-        if solver == "hexaly":
-            time_limit = 5
-        m.solveAll(solver=solver, solution_limit=solution_limit, time_limit=time_limit, display=lambda: sols.add(tuple(argvals(vars))))
+        solution_limit = 15  # ILP solvers don't like this model and tend to get stuck finding all solutions
+        m.solveAll(solver=solver, solution_limit=solution_limit, display=lambda: sols.add(tuple(argvals(vars))))
         for sol in sols:
             xv, yv, dv, rv = sol
             assert dv * yv + rv == xv
@@ -1082,16 +1077,16 @@ def test_status(self, solver):
 
         # now making a tricky problem to solve
         np.random.seed(0)
-        start = cp.intvar(0,100, shape=50)
-        dur = np.random.randint(1,5, size=50)
-        end = cp.intvar(0,100, shape=50)
-        demand  = np.random.randint(10,15, size=50)
+        start = cp.intvar(0,50, shape=20)
+        dur = np.random.randint(1,5, size=20)
+        end = cp.intvar(0,50, shape=20)
+        demand  = np.random.randint(10,15, size=20)
 
-        m += cp.Cumulative(start, dur, end,demand, 30)
+        m += cp.Cumulative(start, dur, end,demand, 20)
         m.minimize(cp.max(end))
         m.solve(solver=solver, time_limit=1)
         # normally, should not be able to solve within 1s...
-        assert m.status().exitstatus in (ExitStatus.FEASIBLE, ExitStatus.UNKNOWN)
+        assert m.status().exitstatus in (ExitStatus.OPTIMAL, ExitStatus.FEASIBLE, ExitStatus.UNKNOWN)
 
         # now trivally unsat
         m += cp.sum(bv) <= 0
@@ -1208,3 +1203,24 @@ def test_objective_numexprs(solver, constraint):
         assert constraint.value() > constraint.get_bounds()[0] # bounds are not always tight, but should be larger than lb for sure
     except NotSupportedError:
         pytest.skip(reason=f"{solver} does not support optimisation")
+
+
+@pytest.mark.skipif(not CPM_scip.supported(), reason="Scip not installed")
+def test_scip_special_cardinality():
+    bvs = cp.boolvar(shape=4)
+    sos1 = cp.sum(bvs) <= 1
+
+    model = cp.Model(sos1)
+    s = cp.SolverLookup.get("scip", model)
+    constraints = s.scip_model.getConss()
+    assert constraints[0].getConshdlrName() == "SOS1"  # translated to native SOS1
+    assert s.solve()
+    assert bvs.value().sum() <= 1
+
+    card = cp.sum(bvs) <= 3
+    model = cp.Model(card)
+    s = cp.SolverLookup.get("scip", model)
+    constraints = s.scip_model.getConss()
+    assert constraints[0].getConshdlrName() == "cardinality"  # translated to native cardinality
+    assert s.solve()
+    assert bvs.value().sum() <= 3