Replace self._curr.model with self._curr.fitted_model (facebook#2019)

Summary:

We want to use the GenerationNode equivalent of model which is fitted_model instead of the GenerationStep field model

Reviewed By: lena-kashtelyan

Differential Revision: D51431892

ax/modelbridge/generation_node.py

@@ -170,6 +170,15 @@ def fitted_model(self) -> ModelBridge:
         """fitted_model from self.model_spec_to_gen_from for convenience"""
         return self.model_spec_to_gen_from.fitted_model
 
+    @property
+    def _fitted_model(self) -> Optional[ModelBridge]:
+        """Private property to return optional fitted_model from
+        self.model_spec_to_gen_from for convenience. If no model is fit,
+        will return None. If using the non-private `fitted_model` property,
+        and no model is fit, a UserInput error will be raised.
+        """
+        return self.model_spec_to_gen_from._fitted_model
+
     @property
     def fixed_features(self) -> Optional[ObservationFeatures]:
         """fixed_features from self.model_spec_to_gen_from for convenience"""

ax/modelbridge/generation_strategy.py

@@ -159,7 +159,7 @@ def model(self) -> Optional[ModelBridge]:
         """Current model in this strategy. Returns None if no model has been set
         yet (i.e., if no generator runs have been produced from this GS).
         """
-        return self._curr.model_spec._fitted_model
+        return self._curr._fitted_model
 
     @property
     def experiment(self) -> Experiment:
@@ -456,7 +456,7 @@ def _fit_current_model(self, data: Optional[Data]) -> None:
             logger.debug(f"Fitting model with data for trials: {trial_indices_in_data}")
 
         self._curr.fit(experiment=self.experiment, data=data, **model_state_on_lgr)
-        self._model = self._curr.model_spec.fitted_model
+        self._model = self._curr._fitted_model
 
     def _maybe_move_to_next_step(self, raise_data_required_error: bool = True) -> bool:
         """Moves this generation strategy to next step if the current step is completed,
@@ -507,7 +507,7 @@ def _get_model_state_from_last_generator_run(self) -> Dict[str, Any]:
         # Potential solution: store generator runs on `GenerationNode`-s and
         # split them per-model there.
         model_state_on_lgr = {}
-        model_on_curr = self._curr.model
+        model_on_curr = self._curr.model_enum
         if (
             lgr is not None
             and lgr._generation_step_index == self._curr.index

ax/storage/json_store/tests/test_json_store.py

@@ -317,6 +317,11 @@ def test_EncodeDecode(self) -> None:
 
                 original_object.evaluation_function = None
                 converted_object.evaluation_function = None
+            if class_ == "BenchmarkMethod":
+                # Some fields of the reloaded GS are not expected to be set (both will
+                # be set during next model fitting call), so we unset them on the
+                # original GS as well.
+                original_object.generation_strategy._unset_non_persistent_state_fields()
             if isinstance(original_object, torch.nn.Module):
                 self.assertIsInstance(
                     converted_object,
@@ -407,6 +412,10 @@ def test_DecodeGenerationStrategy(self) -> None:
             decoder_registry=CORE_DECODER_REGISTRY,
             class_decoder_registry=CORE_CLASS_DECODER_REGISTRY,
         )
+        # Some fields of the reloaded GS are not expected to be set (both will be
+        # set during next model fitting call), so we unset them on the original GS as
+        # well.
+        generation_strategy._unset_non_persistent_state_fields()
         self.assertEqual(generation_strategy, new_generation_strategy)
         self.assertGreater(len(new_generation_strategy._steps), 0)
         self.assertIsInstance(new_generation_strategy._steps[0].model, Models)