implement batch constraint for doe

bofire/strategies/doe/utils.py

@@ -8,6 +8,7 @@
 from scipy.optimize import LinearConstraint, NonlinearConstraint
 
 from bofire.data_models.constraints.api import (
+    InterpointEqualityConstraint,
     LinearEqualityConstraint,
     LinearInequalityConstraint,
     NChooseKConstraint,
@@ -182,7 +183,7 @@ def n_zero_eigvals(
     return len(eigvals) - len(eigvals[eigvals > epsilon])
 
 
-def constraints_as_scipy_constraints(
+def constraints_as_scipy_constraints(  # noqa: C901
     domain: Domain,
     n_experiments: int,
     ignore_nchoosek: bool = True,
@@ -290,7 +291,31 @@ def constraints_as_scipy_constraints(
                 )
 
                 constraints.append(NonlinearConstraint(fun, lb, ub, jac=fun.jacobian))
+        elif isinstance(c, InterpointEqualityConstraint):
+            # similar to linear constraint but with diferent A
+            # write lower/upper bound as vector
+            lb = np.zeros(n_experiments)
+            ub = np.zeros(n_experiments)
 
+            A = np.zeros(shape=(n_experiments, D * n_experiments))
+            multiplicity = c.multiplicity or n_experiments
+            n_max_batch_reps = int(np.ceil(n_experiments / multiplicity))
+            for i in range(n_max_batch_reps):
+                dummy = i * multiplicity
+                for j, p in enumerate(domain.inputs.get_keys()):
+                    if p in c.feature:
+                        # temp_lb = i * multiplicity + 1
+                        # temp_ub = min((i + 1) * multiplicity, n_experiments)
+                        temp_lb = int(dummy + 1)
+                        temp_ub = int(min(dummy + multiplicity, n_experiments))
+                        for k in range(
+                            temp_lb,
+                            temp_ub,
+                        ):
+                            A[k - 1, (k - 1) * D + j] = 1
+                            A[k - 1, k * D + j] = -A[k - 1, (k - 1) * D + j]
+
+            constraints.append(LinearConstraint(A, lb, ub))  # type: ignore
         else:
             raise NotImplementedError(f"No implementation for this constraint: {c}")
 

tests/bofire/strategies/doe/test_design.py

@@ -5,6 +5,7 @@
 import pytest
 
 from bofire.data_models.constraints.api import (
+    InterpointEqualityConstraint,
     LinearEqualityConstraint,
     LinearInequalityConstraint,
     NChooseKConstraint,
@@ -26,6 +27,23 @@
 CYIPOPT_AVAILABLE = importlib.util.find_spec("cyipopt") is not None
 
 
+def test_interpoint_constraint():
+    inputs = [
+        ContinuousInput(
+            key=f"x{i+1}",
+            bounds=(0, 1),
+        )
+        for i in range(4)
+    ]
+    domain = Domain.from_lists(
+        inputs=inputs,
+        outputs=[ContinuousOutput(key="y")],
+        constraints=[InterpointEqualityConstraint(feature="x1", multiplicity=2)],
+    )
+    find_local_max_ipopt(domain, "linear")
+    assert True
+
+
 @pytest.mark.skipif(CYIPOPT_AVAILABLE, reason="requires cyipopt")
 def test_raise_error_if_cyipopt_not_available():
     pytest.raises(ImportError)