Remove deprecated torch_dtype input from Adapters (facebook#3410)

Summary:
Pull Request resolved: facebook#3410

Removes deprecated `torch_dtype` argument from `Adapter` constructors. Updates the storage code to discard the deprecated kwargs to avoid potential errors when loading old experiments.

Reviewed By: dme65

Differential Revision: D69994060

fbshipit-source-id: 2f2d181d588a0727107d6c53d83700ffb9815a0e

ax/benchmark/benchmark_test_functions/surrogate.py

@@ -66,7 +66,7 @@ def evaluate_true(self, params: Mapping[str, TParamValue]) -> Tensor:
         return torch.tensor(
             means,
             device=self.surrogate.device,
-            dtype=self.surrogate.dtype,
+            dtype=torch.double,
         )
 
     @equality_typechecker

ax/modelbridge/factory.py

@@ -108,7 +108,6 @@ def get_botorch(
     experiment: Experiment,
     data: Data,
     search_space: SearchSpace | None = None,
-    dtype: torch.dtype = torch.double,
     device: torch.device = DEFAULT_TORCH_DEVICE,
     transforms: list[type[Transform]] = Cont_X_trans + Y_trans,
     transform_configs: dict[str, TConfig] | None = None,
@@ -127,7 +126,6 @@ def get_botorch(
             experiment=experiment,
             data=data,
             search_space=search_space or experiment.search_space,
-            torch_dtype=dtype,
             torch_device=device,
             transforms=transforms,
             transform_configs=transform_configs,

ax/modelbridge/map_torch.py

@@ -63,7 +63,6 @@ def __init__(
         model: TorchGenerator,
         transforms: list[type[Transform]],
         transform_configs: dict[str, TConfig] | None = None,
-        torch_dtype: torch.dtype | None = None,
         torch_device: torch.device | None = None,
         status_quo_name: str | None = None,
         status_quo_features: ObservationFeatures | None = None,
@@ -90,7 +89,6 @@ def __init__(
                 the reverse order.
             transform_configs: A dictionary from transform name to the
                 transform config dictionary.
-            torch_dtype: Torch data type.
             torch_device: Torch device.
             status_quo_name: Name of the status quo arm. Can only be used if
                 Data has a single set of ObservationFeatures corresponding to
@@ -134,7 +132,6 @@ def __init__(
             model=model,
             transforms=transforms,
             transform_configs=transform_configs,
-            torch_dtype=torch_dtype,
             torch_device=torch_device,
             status_quo_name=status_quo_name,
             status_quo_features=status_quo_features,

ax/modelbridge/tests/test_registry.py

@@ -159,7 +159,6 @@ def test_enum_sobol_legacy_GPEI(self) -> None:
             gpei._bridge_kwargs,
             {
                 "transform_configs": None,
-                "torch_dtype": None,
                 "torch_device": None,
                 "status_quo_name": None,
                 "status_quo_features": None,

ax/modelbridge/tests/test_torch_modelbridge.py

@@ -88,11 +88,9 @@ def test_TorchAdapter(self, device: torch.device | None = None) -> None:
             device=device,
             fit_on_init=False,
         )
-        dtype = torch.double
-        self.assertEqual(model_bridge.dtype, dtype)
         self.assertEqual(model_bridge.device, device)
         self.assertIsNone(model_bridge._last_observations)
-        tkwargs: dict[str, Any] = {"dtype": dtype, "device": device}
+        tkwargs: dict[str, Any] = {"dtype": torch.double, "device": device}
         # Test `_fit`.
         X = torch.tensor([[1.0, 2.0, 3.0], [2.0, 3.0, 4.0]], **tkwargs)
         datasets = {
@@ -285,34 +283,6 @@ def test_TorchAdapter(self, device: torch.device | None = None) -> None:
         X = model_bridge._transform_observation_features(observation_features=obsf)
         self.assertTrue(torch.equal(X, torch.tensor([[1.0, 2.0]], **tkwargs)))
 
-    def _test_TorchAdapter_torch_dtype_deprecated(
-        self, torch_dtype: torch.dtype
-    ) -> None:
-        search_space = get_search_space_for_range_values(
-            min=0.0, max=5.0, parameter_names=["x1", "x2", "x3"]
-        )
-        model = mock.MagicMock(TorchGenerator, autospec=True, instance=True)
-        experiment = Experiment(search_space=search_space, name="test")
-        with self.assertWarnsRegex(
-            DeprecationWarning,
-            "The `torch_dtype` argument to `TorchAdapter` is deprecated",
-        ):
-            TorchAdapter(
-                experiment=experiment,
-                search_space=search_space,
-                data=experiment.lookup_data(),
-                model=model,
-                transforms=[],
-                fit_on_init=False,
-                torch_dtype=torch_dtype,
-            )
-
-    def test_TorchAdapter_float(self) -> None:
-        self._test_TorchAdapter_torch_dtype_deprecated(torch_dtype=torch.float32)
-
-    def test_TorchAdapter_float64(self) -> None:
-        self._test_TorchAdapter_torch_dtype_deprecated(torch_dtype=torch.float64)
-
     def test_TorchAdapter_cuda(self) -> None:
         if torch.cuda.is_available():
             self.test_TorchAdapter(device=torch.device("cuda"))

ax/modelbridge/torch.py

@@ -13,7 +13,6 @@
 from copy import deepcopy
 from logging import Logger
 from typing import Any
-from warnings import warn
 
 import numpy as np
 import numpy.typing as npt
@@ -108,7 +107,6 @@ def __init__(
         model: TorchGenerator,
         transforms: list[type[Transform]],
         transform_configs: dict[str, TConfig] | None = None,
-        torch_dtype: torch.dtype | None = None,
         torch_device: torch.device | None = None,
         status_quo_name: str | None = None,
         status_quo_features: ObservationFeatures | None = None,
@@ -121,19 +119,6 @@ def __init__(
         default_model_gen_options: TConfig | None = None,
         fit_only_completed_map_metrics: bool = True,
     ) -> None:
-        # This warning is being added while we are on 0.4.3, so it will be
-        # released in 0.4.4 or 0.5.0. The `torch_dtype` argument can be removed
-        # in the subsequent minor version. It should also be removed from
-        # `TorchAdapter` subclasses.
-        if torch_dtype is not None:
-            warn(
-                "The `torch_dtype` argument to `TorchAdapter` is deprecated"
-                " and will be ignored; data will be in double precision.",
-                DeprecationWarning,
-            )
-
-        # Note: When `torch_dtype` is removed, this attribute can be removed
-        self.dtype: torch.dtype = torch.double
         self.device = torch_device
         # pyre-ignore [4]: Attribute `_default_model_gen_options` of class
         # `TorchAdapter` must have a type that does not contain `Any`.
@@ -311,7 +296,7 @@ def _array_list_to_tensors(self, arrays: list[npt.NDArray]) -> list[Tensor]:
         return [self._array_to_tensor(x) for x in arrays]
 
     def _array_to_tensor(self, array: npt.NDArray | list[float]) -> Tensor:
-        return torch.as_tensor(array, dtype=self.dtype, device=self.device)
+        return torch.as_tensor(array, dtype=torch.double, device=self.device)
 
     def _convert_observations(
         self,
@@ -367,10 +352,10 @@ def _convert_observations(
                 raise ValueError(f"Outcome `{outcome}` was not observed.")
             X = torch.stack(Xs[outcome], dim=0)
             Y = torch.tensor(
-                Ys[outcome], dtype=self.dtype, device=self.device
+                Ys[outcome], dtype=torch.double, device=self.device
             ).unsqueeze(-1)
             Yvar = torch.tensor(
-                Yvars[outcome], dtype=self.dtype, device=self.device
+                Yvars[outcome], dtype=torch.double, device=self.device
             ).unsqueeze(-1)
             if Yvar.isnan().all():
                 Yvar = None
@@ -468,13 +453,13 @@ def _cross_validate(
             parameters = self.parameters
         X_test = torch.tensor(
             [[obsf.parameters[p] for p in parameters] for obsf in cv_test_points],
-            dtype=self.dtype,
+            dtype=torch.double,
             device=self.device,
         )
         # Use the model to do the cross validation
         f_test, cov_test = none_throws(self.model).cross_validate(
             datasets=datasets,
-            X_test=torch.as_tensor(X_test, dtype=self.dtype, device=self.device),
+            X_test=torch.as_tensor(X_test, dtype=torch.double, device=self.device),
             search_space_digest=search_space_digest,
             use_posterior_predictive=use_posterior_predictive,
         )
@@ -1040,7 +1025,7 @@ def _extract_observation_data(
             try:
                 x = torch.tensor(
                     [obsf.parameters[p] for p in parameters],
-                    dtype=self.dtype,
+                    dtype=torch.double,
                     device=self.device,
                 )
             except (KeyError, TypeError):

ax/storage/json_store/decoder.py

@@ -85,6 +85,9 @@
     "ST_MTGP_NEHVI": "ST_MTGP",
 }
 
+# Deprecated model kwargs, to be removed from GStep / GNodes.
+_DEPRECATED_MODEL_KWARGS: tuple[str, ...] = ("fit_on_update", "torch_dtype")
+
 
 @dataclass
 class RegistryKwargs:
@@ -727,7 +730,9 @@ def generation_step_from_json(
         generation_step_json
     )
     kwargs = generation_step_json.pop("model_kwargs", None)
-    kwargs.pop("fit_on_update", None)  # Remove deprecated fit_on_update.
+    for k in _DEPRECATED_MODEL_KWARGS:
+        # Remove deprecated kwargs.
+        kwargs.pop(k, None)
     if kwargs is not None:
         kwargs = _extract_surrogate_spec_from_surrogate_specs(kwargs)
     gen_kwargs = generation_step_json.pop("model_gen_kwargs", None)
@@ -788,7 +793,9 @@ def model_spec_from_json(
 ) -> GeneratorSpec:
     """Load GeneratorSpec from JSON."""
     kwargs = model_spec_json.pop("model_kwargs", None)
-    kwargs.pop("fit_on_update", None)  # Remove deprecated fit_on_update.
+    for k in _DEPRECATED_MODEL_KWARGS:
+        # Remove deprecated model kwargs.
+        kwargs.pop(k, None)
     if kwargs is not None:
         kwargs = _extract_surrogate_spec_from_surrogate_specs(kwargs)
     gen_kwargs = model_spec_json.pop("model_gen_kwargs", None)

ax/storage/json_store/tests/test_json_store.py

@@ -804,7 +804,11 @@ def test_generation_step_backwards_compatibility(self) -> None:
             "max_parallelism": None,
             "use_update": False,
             "enforce_num_trials": True,
-            "model_kwargs": {"fit_on_update": False, "other_kwarg": 5},
+            "model_kwargs": {
+                "fit_on_update": False,
+                "torch_dtype": torch.double,
+                "other_kwarg": 5,
+            },
             "model_gen_kwargs": {},
             "index": -1,
             "should_deduplicate": False,

ax/utils/testing/core_stubs.py

@@ -1905,7 +1905,7 @@ def get_generator_run() -> GeneratorRun:
         gen_time=5.0,
         model_key="Sobol",
         model_kwargs={"scramble": False, "torch_device": torch.device("cpu")},
-        bridge_kwargs={"transforms": Cont_X_trans, "torch_dtype": torch.double},
+        bridge_kwargs={"transforms": Cont_X_trans},
         generation_step_index=0,
         candidate_metadata_by_arm_signature={
             a.signature: {"md_key": f"md_val_{a.signature}"} for a in arms

tutorials/multi_task/multi_task.ipynb

@@ -370,7 +370,6 @@
     "    search_space: Optional[SearchSpace] = None,\n",
     "    trial_index: Optional[int] = None,\n",
     "    device: torch.device = torch.device(\"cpu\"),\n",
-    "    dtype: torch.dtype = torch.double,\n",
     ") -> TorchAdapter:\n",
     "    \"\"\"Instantiates a Multi-task Gaussian Process (MTGP) model that generates\n",
     "    points with EI.\n",
@@ -419,7 +418,6 @@
     "            data=data,\n",
     "            transforms=transforms,\n",
     "            transform_configs=transform_configs,\n",
-    "            torch_dtype=dtype,\n",
     "            torch_device=device,\n",
     "            status_quo_features=status_quo_features,\n",
     "        ),\n",