Add log_scale option to ChoiceParameterConfig

ax/api/configs.py

@@ -40,6 +40,7 @@ class ChoiceParameterConfig:
     values: list[float] | list[int] | list[str] | list[bool]
     parameter_type: Literal["float", "int", "str", "bool"]
     is_ordered: bool | None = None
+    log_scale: bool | None = None
     dependent_parameters: Mapping[TParameterValue, Sequence[str]] | None = None
 
 

ax/api/utils/instantiation/from_config.py

@@ -93,6 +93,7 @@ def parameter_from_config(
             parameter_type=_parameter_type_converter(config.parameter_type),
             values=cast(list[TParamValue], config.values),
             is_ordered=config.is_ordered,
+            log_scale=config.log_scale,
             dependents=cast(
                 dict[TParamValue, list[str]] | None,
                 config.dependent_parameters,

ax/api/utils/instantiation/tests/test_from_config.py

@@ -26,6 +26,7 @@
 from ax.core.search_space import SearchSpace
 from ax.exceptions.core import UserInputError
 from ax.utils.common.testutils import TestCase
+from pyre_extensions import assert_is_instance
 
 
 class TestFromConfig(TestCase):
@@ -222,6 +223,46 @@ def test_create_choice_parameter(self) -> None:
                 value="a",
             ),
         )
+
+        choice_config_with_log_scale = ChoiceParameterConfig(
+            name="choice_param_with_log_scale",
+            parameter_type="float",
+            values=[1.0, 2.0, 3.0],
+            log_scale=True,
+        )
+        self.assertEqual(
+            parameter_from_config(config=choice_config_with_log_scale),
+            ChoiceParameter(
+                name="choice_param_with_log_scale",
+                parameter_type=CoreParameterType.FLOAT,
+                values=[1.0, 2.0, 3.0],
+                log_scale=True,
+            ),
+        )
+
+        # log_scale=False should override the auto-detection default.
+        choice_config_without_log_scale = ChoiceParameterConfig(
+            name="choice_param_without_log_scale",
+            parameter_type="int",
+            values=[1, 10, 100],
+            log_scale=False,
+        )
+        parameter = parameter_from_config(config=choice_config_without_log_scale)
+        self.assertFalse(assert_is_instance(parameter, ChoiceParameter).log_scale)
+
+        # log_scale is only supported for numeric parameters.
+        with self.assertRaisesRegex(
+            UserInputError, "log_scale is only supported for numerical parameters"
+        ):
+            parameter_from_config(
+                config=ChoiceParameterConfig(
+                    name="str_choice_param_with_log_scale",
+                    parameter_type="str",
+                    values=["a", "b", "c"],
+                    log_scale=True,
+                )
+            )
+
         self.assertFalse(any("sort_values" in str(w.message) for w in ws))
 
     def test_experiment_from_config(self) -> None: