Unflakify TestOptimizeAcqfMixed.test_continuous_step (#2578)

Summary:
Pull Request resolved: #2578

Due to randomized continuous & binary dimensions, the fixed continuous dimension would occasionally become part of the constraint and fail the test since the fixed value cannot satisfy the constraint. This diff updates the fixed dimension to ensure this cannot happen.

Reviewed By: Balandat

Differential Revision: D64475003

fbshipit-source-id: 03164639e462bc7583c0b1ba47cbeb9463085c4b

Author: saitcakmak

test/optim/test_optimize_acqf_mixed.py

@@ -354,7 +354,7 @@ def test_continuous_step(self):
 
         best_f = model(X)
         ei = ExpectedImprovement(model, best_f=best_f)
-        X_clone, ei_val = continuous_step(
+        X_new, ei_val = continuous_step(
             opt_inputs=_make_opt_inputs(
                 acq_function=ei,
                 bounds=bounds,
@@ -363,15 +363,18 @@ def test_continuous_step(self):
             discrete_dims=binary_dims,
             current_x=X.clone(),
         )
-        self.assertAllClose(X_clone[cont_dims], root[cont_dims])
-        self.assertAllClose(X_clone[binary_dims], X[binary_dims])
+        self.assertAllClose(X_new[cont_dims], root[cont_dims])
+        self.assertAllClose(X_new[binary_dims], X[binary_dims])
 
         # Test with fixed features and constraints.
         fixed_binary = int(binary_dims[0])
-        fixed_cont = int(cont_dims[0])
+        # We don't want fixed cont to be one of the first two indices,
+        # to avoid it being a part of the constraint. This ensures that.
+        # The fixed value of 0.5 cannot satisfy the constraint.
+        fixed_cont = int(cont_dims[:3].max())
         X_ = X.clone()
         X_[:2] = 1.0  # To satisfy the constraint.
-        X_clone, ei_val = continuous_step(
+        X_new, ei_val = continuous_step(
             opt_inputs=_make_opt_inputs(
                 acq_function=ei,
                 bounds=bounds,
@@ -390,11 +393,11 @@ def test_continuous_step(self):
         )
         self.assertTrue(
             torch.equal(
-                X_clone[[fixed_binary, fixed_cont]],
+                X_new[[fixed_binary, fixed_cont]],
                 torch.tensor([1.0, 0.5], device=self.device),
             )
         )
-        self.assertAllClose(X_clone[:2], X_[:2])
+        self.assertAllClose(X_new[:2], X_[:2])
 
         # test edge case when all parameters are binary
         root = torch.rand(d_bin)
@@ -422,7 +425,7 @@ def test_continuous_step(self):
             ValueError,
             "continuous_step requires current_x to be",
         ):
-            X_clone, ei_val = continuous_step(
+            X_new, ei_val = continuous_step(
                 opt_inputs=_make_opt_inputs(
                     acq_function=ei,
                     bounds=bounds,