Update nonlinear_constraint_is_feasible to return a boolean tensor (#2731)

Summary:
Pull Request resolved: #2731

`nonlinear_constraint_is_feasible` checks whether the constraint is feasible for the given (batch of) candidates. Previously, this returned a boolean, which was False if any element of the batch was infeasible. This diff updates it to return a boolean tensor that  shows whether each batch is feasible. Use cases are updated to comply with the new behavior.

I'll utilize this in a follow up diff to introduce a helper that evaluates the feasibility of all forms of constraints.

Reviewed By: dme65

Differential Revision: D69209007

fbshipit-source-id: 4f63ee6824734c3eb5c96bb5b1cf47a17d51b298

Author: saitcakmak

botorch/generation/gen.py

@@ -265,17 +265,21 @@ def f(x):
     # so it shouldn't be an issue given enough restarts.
     if nonlinear_inequality_constraints:
         for con, is_intrapoint in nonlinear_inequality_constraints:
-            if not nonlinear_constraint_is_feasible(
-                con, is_intrapoint=is_intrapoint, x=candidates
-            ):
-                candidates = torch.from_numpy(x0).to(candidates).reshape(shapeX)
+            if not (
+                feasible := nonlinear_constraint_is_feasible(
+                    con, is_intrapoint=is_intrapoint, x=candidates
+                )
+            ).all():
+                # Replace the infeasible batches with feasible ICs.
+                candidates[~feasible] = (
+                    torch.from_numpy(x0).to(candidates).reshape(shapeX)[~feasible]
+                )
                 warnings.warn(
                     "SLSQP failed to converge to a solution the satisfies the "
                     "non-linear constraints. Returning the feasible starting point.",
                     OptimizationWarning,
                     stacklevel=2,
                 )
-                break
 
     clamped_candidates = columnwise_clamp(
         X=candidates, lower=lower_bounds, upper=upper_bounds, raise_on_violation=True

botorch/optim/parameter_constraints.py

@@ -512,7 +512,7 @@ def f_grad(X):
 
 def nonlinear_constraint_is_feasible(
     nonlinear_inequality_constraint: Callable, is_intrapoint: bool, x: Tensor
-) -> bool:
+) -> Tensor:
     """Checks if a nonlinear inequality constraint is fulfilled.
 
     Args:
@@ -522,23 +522,24 @@ def nonlinear_constraint_is_feasible(
             is applied pointwise and is broadcasted over the q-batch. Else, the
             constraint has to evaluated over the whole q-batch and is a an
             inter-point constraint.
-        x: Tensor of shape (b x q x d).
+        x: Tensor of shape (batch x q x d).
 
     Returns:
-        bool: True if the constraint is fulfilled, else False.
+        A boolean tensor of shape (batch) indicating if the constraint is
+        satified by the corresponding batch of `x`.
     """
 
     def check_x(x: Tensor) -> bool:
         return _arrayify(nonlinear_inequality_constraint(x)).item() >= NLC_TOL
 
-    for x_ in x:
+    x_flat = x.view(-1, *x.shape[-2:])
+    is_feasible = torch.ones(x_flat.shape[0], dtype=torch.bool, device=x.device)
+    for i, x_ in enumerate(x_flat):
         if is_intrapoint:
-            if not all(check_x(x__) for x__ in x_):
-                return False
+            is_feasible[i] &= all(check_x(x__) for x__ in x_)
         else:
-            if not check_x(x_):
-                return False
-    return True
+            is_feasible[i] &= check_x(x_)
+    return is_feasible.view(x.shape[:-2])
 
 
 def make_scipy_nonlinear_inequality_constraints(
@@ -589,7 +590,7 @@ def make_scipy_nonlinear_inequality_constraints(
         nlc, is_intrapoint = constraint
         if not nonlinear_constraint_is_feasible(
             nlc, is_intrapoint=is_intrapoint, x=x0.reshape(shapeX)
-        ):
+        ).all():
             raise ValueError(
                 "`batch_initial_conditions` must satisfy the non-linear inequality "
                 "constraints."

test/optim/test_parameter_constraints.py

@@ -358,15 +358,16 @@ def nlc(x):
                 ),
             )
         )
-        self.assertFalse(
+        self.assertEqual(
             nonlinear_constraint_is_feasible(
                 nlc,
                 True,
                 torch.tensor(
                     [[[1.5, 1.5], [1.5, 1.5]], [[1.5, 1.5], [1.5, 3.5]]],
                     device=self.device,
                 ),
-            )
+            ).tolist(),
+            [True, False],
         )
         self.assertTrue(
             nonlinear_constraint_is_feasible(
@@ -381,22 +382,23 @@ def nlc(x):
                     [[[1.0, 1.0], [1.0, 1.0]], [[1.0, 1.0], [1.0, 1.0]]],
                     device=self.device,
                 ),
-            )
+            ).all()
         )
         self.assertFalse(
             nonlinear_constraint_is_feasible(
                 nlc, False, torch.tensor([[[1.5, 1.5], [1.5, 1.5]]], device=self.device)
             )
         )
-        self.assertFalse(
+        self.assertEqual(
             nonlinear_constraint_is_feasible(
                 nlc,
                 False,
                 torch.tensor(
                     [[[1.0, 1.0], [1.0, 1.0]], [[1.5, 1.5], [1.5, 1.5]]],
                     device=self.device,
                 ),
-            )
+            ).tolist(),
+            [True, False],
         )
 
     def test_generate_unfixed_nonlin_constraints(self):