fix X_observed check when constructing objective (#3444)

Summary:
Pull Request resolved: #3444

This check is not needed for SampleReducingMCAcquisitionFunction and blocks generation when there is no X_baseline

Reviewed By: Balandat, SebastianAment

Differential Revision: D70446607

fbshipit-source-id: 12be60bb8e40f27df98a6c92163c8bc163a950e7

ax/models/tests/test_torch_utils.py

@@ -130,6 +130,16 @@ def test_get_botorch_objective(self, _) -> None:
         )
         self.assertIsInstance(obj, GenericMCObjective)
         self.assertIsNone(tf)
+        # test no X_observed (e.g., if there are no feasible points) with qLogNEI
+        obj, tf = get_botorch_objective_and_transform(
+            botorch_acqf_class=qLogNoisyExpectedImprovement,
+            model=self.mock_botorch_model,
+            outcome_constraints=self.outcome_constraints,
+            objective_weights=self.objective_weights,
+            X_observed=None,
+        )
+        self.assertIsInstance(obj, GenericMCObjective)
+        self.assertIsNone(tf)
 
         # By default, `ScalarizedPosteriorTransform` should be picked in absence of
         # outcome constraints.

ax/models/torch/utils.py

@@ -464,10 +464,6 @@ def get_botorch_objective_and_transform(
         # We are doing multi-objective optimization.
         return _get_weighted_mo_objective(objective_weights=objective_weights), None
     if outcome_constraints:
-        if X_observed is None:
-            raise UnsupportedError(
-                "X_observed is required to construct a constrained BoTorch objective."
-            )
         # If there are outcome constraints, we use MC Acquisition functions.
         obj_tf: Callable[[Tensor, Tensor | None], Tensor] = (
             get_objective_weights_transform(objective_weights)
@@ -481,13 +477,17 @@ def objective(samples: Tensor, X: Tensor | None = None) -> Tensor:
         # Acquisition object.
         if issubclass(botorch_acqf_class, SampleReducingMCAcquisitionFunction):
             return GenericMCObjective(objective=objective), None
-        else:  # this is still used by KG
-            con_tfs = get_outcome_constraint_transforms(outcome_constraints)
-            inf_cost = get_infeasible_cost(X=X_observed, model=model, objective=obj_tf)
-            objective = ConstrainedMCObjective(
-                objective=objective, constraints=con_tfs or [], infeasible_cost=inf_cost
+        # this is still used by KG
+        if X_observed is None:
+            raise UnsupportedError(
+                "X_observed is required to construct a constrained BoTorch objective."
             )
-            return objective, None
+        con_tfs = get_outcome_constraint_transforms(outcome_constraints)
+        inf_cost = get_infeasible_cost(X=X_observed, model=model, objective=obj_tf)
+        objective = ConstrainedMCObjective(
+            objective=objective, constraints=con_tfs or [], infeasible_cost=inf_cost
+        )
+        return objective, None
     # Case of linear weights - use ScalarizedPosteriorTransform
     transform = ScalarizedPosteriorTransform(weights=objective_weights)
     return None, transform