Remove parameter transformers from config object state,

composing the list dynamically instead

tianshou/highlevel/params/policy_params.py

@@ -1,6 +1,6 @@
 from abc import ABC, abstractmethod
-from dataclasses import asdict, dataclass, field
-from typing import Any, Literal
+from dataclasses import asdict, dataclass
+from typing import Any, Literal, Protocol
 
 import torch
 
@@ -159,39 +159,33 @@ class ParamTransformerFloatEnvParamFactory(ParamTransformer):
             kwargs[self.key] = value.create_value(data.envs)
 
 
-class ITransformableParams(ABC):
-    @abstractmethod
-    def _add_transformer(self, transformer: ParamTransformer):
+class GetParamTransformersProtocol(Protocol):
+    def _get_param_transformers(self) -> list[ParamTransformer]:
         pass
 
 
 @dataclass
-class Params(ITransformableParams):
-    _param_transformers: list[ParamTransformer] = field(
-        init=False,
-        default_factory=list,
-        repr=False,
-    )
-
-    def _add_transformer(self, transformer: ParamTransformer):
-        self._param_transformers.append(transformer)
-
+class Params(GetParamTransformersProtocol):
     def create_kwargs(self, data: ParamTransformerData) -> dict[str, Any]:
         params = asdict(self)
-        for transformer in self._param_transformers:
+        for transformer in self._get_param_transformers():
             transformer.transform(params, data)
-        del params["_param_transformers"]
         return params
 
+    def _get_param_transformers(self) -> list[ParamTransformer]:
+        return []
+
 
 @dataclass
-class ParamsMixinLearningRateWithScheduler(ITransformableParams, ABC):
+class ParamsMixinLearningRateWithScheduler(GetParamTransformersProtocol):
     lr: float = 1e-3
     lr_scheduler_factory: LRSchedulerFactory | None = None
 
-    def __post_init__(self):
-        self._add_transformer(ParamTransformerDrop("lr"))
-        self._add_transformer(ParamTransformerLRScheduler("lr_scheduler_factory", "lr_scheduler"))
+    def _get_param_transformers(self):
+        return [
+            ParamTransformerDrop("lr"),
+            ParamTransformerLRScheduler("lr_scheduler_factory", "lr_scheduler"),
+        ]
 
 
 @dataclass
@@ -226,7 +220,7 @@ class PPOParams(A2CParams, ParamsMixinLearningRateWithScheduler):
 
 
 @dataclass
-class ParamsMixinActorAndDualCritics(ITransformableParams, ABC):
+class ParamsMixinActorAndDualCritics(GetParamTransformersProtocol):
     actor_lr: float = 1e-3
     critic1_lr: float = 1e-3
     critic2_lr: float = 1e-3
@@ -234,16 +228,16 @@ class ParamsMixinActorAndDualCritics(ITransformableParams, ABC):
     critic1_lr_scheduler_factory: LRSchedulerFactory | None = None
     critic2_lr_scheduler_factory: LRSchedulerFactory | None = None
 
-    def __post_init__(self):
-        self._add_transformer(ParamTransformerDrop("actor_lr", "critic1_lr", "critic2_lr"))
-        self._add_transformer(
+    def _get_param_transformers(self):
+        return [
+            ParamTransformerDrop("actor_lr", "critic1_lr", "critic2_lr"),
             ParamTransformerActorDualCriticsLRScheduler(
                 "actor_lr_scheduler_factory",
                 "critic1_lr_scheduler_factory",
                 "critic2_lr_scheduler_factory",
                 "lr_scheduler",
             ),
-        )
+        ]
 
 
 @dataclass
@@ -257,10 +251,12 @@ class SACParams(Params, ParamsMixinActorAndDualCritics):
     action_scaling: bool = True
     action_bound_method: Literal["clip"] | None = "clip"
 
-    def __post_init__(self):
-        ParamsMixinActorAndDualCritics.__post_init__(self)
-        self._add_transformer(ParamTransformerAutoAlpha("alpha"))
-        self._add_transformer(ParamTransformerNoiseFactory("exploration_noise"))
+    def _get_param_transformers(self):
+        transformers = super()._get_param_transformers()
+        transformers.extend(ParamsMixinActorAndDualCritics._get_param_transformers(self))
+        transformers.append(ParamTransformerAutoAlpha("alpha"))
+        transformers.append(ParamTransformerNoiseFactory("exploration_noise"))
+        return transformers
 
 
 @dataclass
@@ -275,9 +271,10 @@ class TD3Params(Params, ParamsMixinActorAndDualCritics):
     action_scaling: bool = True
     action_bound_method: Literal["clip"] | None = "clip"
 
-    # TODO change to stateless variant
-    def __post_init__(self):
-        ParamsMixinActorAndDualCritics.__post_init__(self)
-        self._add_transformer(ParamTransformerNoiseFactory("exploration_noise"))
-        self._add_transformer(ParamTransformerFloatEnvParamFactory("policy_noise"))
-        self._add_transformer(ParamTransformerFloatEnvParamFactory("noise_clip"))
+    def _get_param_transformers(self):
+        transformers = super()._get_param_transformers()
+        transformers.extend(ParamsMixinActorAndDualCritics._get_param_transformers(self))
+        transformers.append(ParamTransformerNoiseFactory("exploration_noise"))
+        transformers.append(ParamTransformerFloatEnvParamFactory("policy_noise"))
+        transformers.append(ParamTransformerFloatEnvParamFactory("noise_clip"))
+        return transformers