Fix flaky test_get_best_f_mc test (#1969)

Summary:
Pull Request resolved: #1969

This test started to become flaky recently, presumably b/c to some minor changes in the numerics on GPU computations. Replacing equality with closeness check fixes this.

Reviewed By: saitcakmak

Differential Revision: D48033053

fbshipit-source-id: 5261ec92ca67aef82567c994aeb6ca802a1e07a4

Author: Balandat

test/acquisition/test_input_constructors.py

@@ -140,22 +140,22 @@ def test_get_best_f_mc(self):
         best_f = get_best_f_mc(training_data=self.blockX_blockY)
         self.assertEqual(best_f, get_best_f_mc(self.blockX_blockY[0]))
 
-        best_f_expected = self.blockX_blockY[0].Y().squeeze().max()
-        self.assertEqual(best_f, best_f_expected)
+        best_f_expected = self.blockX_blockY[0].Y().max(dim=0).values
+        self.assertAllClose(best_f, best_f_expected)
         with self.assertRaisesRegex(UnsupportedError, "require an objective"):
             get_best_f_mc(training_data=self.blockX_multiY)
         obj = LinearMCObjective(weights=torch.rand(2))
         best_f = get_best_f_mc(training_data=self.blockX_multiY, objective=obj)
 
         multi_Y = torch.cat([d.Y() for d in self.blockX_multiY.values()], dim=-1)
         best_f_expected = (multi_Y @ obj.weights).amax(dim=-1, keepdim=True)
-        self.assertEqual(best_f, best_f_expected)
+        self.assertAllClose(best_f, best_f_expected)
         post_tf = ScalarizedPosteriorTransform(weights=torch.ones(2))
         best_f = get_best_f_mc(
             training_data=self.blockX_multiY, posterior_transform=post_tf
         )
         best_f_expected = (multi_Y.sum(dim=-1)).amax(dim=-1, keepdim=True)
-        self.assertEqual(best_f, best_f_expected)
+        self.assertAllClose(best_f, best_f_expected)
 
     @mock.patch("botorch.acquisition.input_constructors.optimize_acqf")
     def test_optimize_objective(self, mock_optimize_acqf):