Add test coverage for warning in Model._set_transformed_inputs (#3107)

Summary:
Pull Request resolved: #3107

This commit adds test coverage for the warning in `Model._set_transformed_inputs`. [This is currently lacking coverage](https://app.codecov.io/gh/meta-pytorch/botorch/pull/3103/blob/botorch/models/model.py?dropdown=coverage).

Reviewed By: esantorella

Differential Revision: D88653645

fbshipit-source-id: 2b3831635711cc31816250741e494a1502b1cb8b

Author: SebastianAment

test/models/test_model.py

@@ -10,6 +10,7 @@
 from botorch.exceptions.errors import InputDataError
 from botorch.models.deterministic import GenericDeterministicModel
 from botorch.models.model import Model, ModelDict, ModelList
+from botorch.models.transforms.input import Normalize
 from botorch.posteriors.ensemble import EnsemblePosterior
 from botorch.posteriors.posterior_list import PosteriorList
 from botorch.utils.datasets import SupervisedDataset
@@ -22,6 +23,22 @@ def posterior(self, X, output_indices, observation_noise, **kwargs):
         pass
 
 
+class ModelWithInputTransformButNoTrainInputs(Model):
+    """A model that has input_transform but no train_inputs attribute."""
+
+    def __init__(self, input_transform):
+        """Initialize the model with an input transform.
+
+        Args:
+            input_transform: The input transform to apply.
+        """
+        super().__init__()
+        self.input_transform = input_transform
+
+    def posterior(self, X, output_indices, observation_noise, **kwargs):
+        pass
+
+
 class GenericDeterministicModelWithBatchShape(GenericDeterministicModel):
     # mocking torch.nn.Module components is kind of funky, so let's do this instead
     @property
@@ -58,6 +75,33 @@ def test_not_so_abstract_base_model(self):
         with self.assertRaises(NotImplementedError):
             model.subset_output([0])
 
+    def test_set_transformed_inputs_warning_without_train_inputs(self) -> None:
+        # Test that a RuntimeWarning is raised when a model has an input_transform
+        # but no train_inputs attribute.
+        input_transform = Normalize(d=2)
+        model = ModelWithInputTransformButNoTrainInputs(input_transform=input_transform)
+
+        # Verify the model has input_transform but no train_inputs
+        self.assertTrue(hasattr(model, "input_transform"))
+        self.assertFalse(hasattr(model, "train_inputs"))
+
+        # Test cases: (method_name, callable that triggers _set_transformed_inputs)
+        test_cases = [
+            ("_set_transformed_inputs", lambda: model._set_transformed_inputs()),
+            ("eval", lambda: model.eval()),
+            ("train(mode=False)", lambda: model.train(mode=False)),
+        ]
+
+        for method_name, trigger_fn in test_cases:
+            with self.subTest(method=method_name):
+                with self.assertWarnsRegex(
+                    RuntimeWarning,
+                    "Could not update `train_inputs` with transformed inputs since "
+                    "ModelWithInputTransformButNoTrainInputs does not have a "
+                    "`train_inputs` attribute",
+                ):
+                    trigger_fn()
+
     def test_construct_inputs(self) -> None:
         model = NotSoAbstractBaseModel()
         with self.subTest("Wrong training data type"), self.assertRaisesRegex(