Deprecate Models.GPEI registry entry (#3020)

Summary:
Pull Request resolved: #3020

This diff deprecates `Models.GPEI`, which has been the default model used in many places in the past. I cleaned up all usage I could find, and updated `Models.GPEI` to point to `Models.BOTORCH_MODULAR` with a deprecation warning. At a later date, we can clean that up with a storage level change to support backwards compatibility to ensure we can continue loading old experiments.

Reviewed By: Balandat

Differential Revision: D64987622

fbshipit-source-id: 84a63ee07cb148aaa706efdf832be4b5e04c1ad8

Author: saitcakmak

ax/analysis/plotly/tests/test_insample_effects.py

@@ -39,12 +39,12 @@ def setUp(self) -> None:
                     transition_criteria=[
                         MaxTrials(
                             threshold=1,
-                            transition_to="GPEI",
+                            transition_to="MBM",
                         )
                     ],
                 ),
                 GenerationNode(
-                    node_name="GPEI",
+                    node_name="MBM",
                     model_specs=[
                         ModelSpec(
                             model_enum=Models.BOTORCH_MODULAR,

ax/analysis/plotly/tests/test_predicted_effects.py

@@ -44,12 +44,12 @@ def setUp(self) -> None:
                     transition_criteria=[
                         MaxTrials(
                             threshold=1,
-                            transition_to="GPEI",
+                            transition_to="MBM",
                         )
                     ],
                 ),
                 GenerationNode(
-                    node_name="GPEI",
+                    node_name="MBM",
                     model_specs=[
                         ModelSpec(
                             model_enum=Models.BOTORCH_MODULAR,

ax/modelbridge/registry.py

@@ -159,7 +159,7 @@ class ModelSetup(NamedTuple):
         transforms=Cont_X_trans + Y_trans,
         standard_bridge_kwargs=STANDARD_TORCH_BRIDGE_KWARGS,
     ),
-    "GPEI": ModelSetup(
+    "Legacy_GPEI": ModelSetup(
         bridge_class=TorchModelBridge,
         model_class=BotorchModel,
         transforms=Cont_X_trans + Y_trans,
@@ -427,12 +427,11 @@ class Models(ModelRegistryBase):
     """
 
     SOBOL = "Sobol"
-    GPEI = "GPEI"
     FACTORIAL = "Factorial"
     SAASBO = "SAASBO"
     SAAS_MTGP = "SAAS_MTGP"
     THOMPSON = "Thompson"
-    LEGACY_BOTORCH = "GPEI"
+    LEGACY_BOTORCH = "Legacy_GPEI"
     BOTORCH_MODULAR = "BoTorch"
     EMPIRICAL_BAYES_THOMPSON = "EB"
     UNIFORM = "Uniform"
@@ -443,6 +442,13 @@ class Models(ModelRegistryBase):
     ST_MTGP_NEHVI = "ST_MTGP_NEHVI"
     CONTEXT_SACBO = "Contextual_SACBO"
 
+    @classmethod
+    @property
+    def GPEI(cls) -> Models:
+        return _deprecated_model_with_warning(
+            old_model_str="GPEI", new_model=cls.BOTORCH_MODULAR
+        )
+
     @classmethod
     @property
     def FULLYBAYESIAN(cls) -> Models:

ax/modelbridge/tests/test_registry.py

@@ -103,8 +103,8 @@ def test_SAASBO(self) -> None:
         )
 
     @mock_botorch_optimize
-    def test_enum_sobol_GPEI(self) -> None:
-        """Tests Sobol and GPEI instantiation through the Models enum."""
+    def test_enum_sobol_legacy_GPEI(self) -> None:
+        """Tests Sobol and Legacy GPEI instantiation through the Models enum."""
         exp = get_branin_experiment()
         # Check that factory generates a valid sobol modelbridge.
         sobol = Models.SOBOL(search_space=exp.search_space)
@@ -115,9 +115,9 @@ def test_enum_sobol_GPEI(self) -> None:
             exp.new_batch_trial().add_generator_run(sobol_run).run()
         # Check that factory generates a valid GP+EI modelbridge.
         exp.optimization_config = get_branin_optimization_config()
-        gpei = Models.GPEI(experiment=exp, data=exp.fetch_data())
+        gpei = Models.LEGACY_BOTORCH(experiment=exp, data=exp.fetch_data())
         self.assertIsInstance(gpei, TorchModelBridge)
-        self.assertEqual(gpei._model_key, "GPEI")
+        self.assertEqual(gpei._model_key, "Legacy_GPEI")
         botorch_defaults = "ax.models.torch.botorch_defaults"
         # Check that the callable kwargs and the torch kwargs were recorded.
         self.assertEqual(
@@ -168,7 +168,7 @@ def test_enum_sobol_GPEI(self) -> None:
             },
         )
         prior_kwargs = {"lengthscale_prior": GammaPrior(6.0, 6.0)}
-        gpei = Models.GPEI(
+        gpei = Models.LEGACY_BOTORCH(
             experiment=exp,
             data=exp.fetch_data(),
             search_space=exp.search_space,
@@ -316,10 +316,10 @@ def test_get_model_from_generator_run(self) -> None:
         self.assertEqual(initial_sobol.gen(n=1).arms, sobol_after_gen.gen(n=1).arms)
         exp.new_trial(generator_run=gr)
         # Check restoration of GPEI, to ensure proper restoration of callable kwargs
-        gpei = Models.GPEI(experiment=exp, data=get_branin_data())
+        gpei = Models.LEGACY_BOTORCH(experiment=exp, data=get_branin_data())
         # Punch GPEI model + bridge kwargs into the Sobol generator run, to avoid
         # a slow call to `gpei.gen`, and remove Sobol's model state.
-        gr._model_key = "GPEI"
+        gr._model_key = "Legacy_GPEI"
         gr._model_kwargs = gpei._model_kwargs
         gr._bridge_kwargs = gpei._bridge_kwargs
         gr._model_state_after_gen = {}
@@ -496,6 +496,7 @@ def test_deprecated_models(self) -> None:
         same check in a couple different ways.
         """
         for old_model_str, new_model in [
+            ("GPEI", Models.BOTORCH_MODULAR),
             ("FULLYBAYESIAN", Models.SAASBO),
             ("FULLYBAYESIANMOO", Models.SAASBO),
             ("FULLYBAYESIAN_MTGP", Models.SAAS_MTGP),

ax/modelbridge/tests/test_robust_modelbridge.py

@@ -129,7 +129,7 @@ def test_mars(self) -> None:
     def test_unsupported_model(self) -> None:
         exp = get_robust_branin_experiment()
         with self.assertRaisesRegex(UnsupportedError, "support robust"):
-            Models.GPEI(
+            Models.LEGACY_BOTORCH(
                 experiment=exp,
                 data=exp.fetch_data(),
             ).gen(n=1)

ax/service/ax_client.py

@@ -157,7 +157,7 @@ class AxClient(WithDBSettingsBase, BestPointMixin, InstantiationBase):
 
         torch_device: An optional `torch.device` object, used to choose the device
             used for generating new points for trials. Works only for torch-based
-            models, such as GPEI. Ignored if a `generation_strategy` is passed in
+            models, such as MBM. Ignored if a `generation_strategy` is passed in
             manually. To specify the device for a custom `generation_strategy`,
             pass in `torch_device` as part of `model_kwargs`. See
             https://ax.dev/tutorials/generation_strategy.html for a tutorial on
@@ -1119,8 +1119,7 @@ def get_feature_importances(self, relative: bool = True) -> AxPlotConfig:
                 logger.info(
                     f"Model {self.generation_strategy.model} does not implement "
                     "`feature_importances`, so it cannot be used to generate "
-                    "this plot. Only certain models, specifically GPEI, implement "
-                    "feature importances."
+                    "this plot. Only certain models, implement feature importances."
                 )
 
         raise ValueError(