[RLlib] Remove all default config objects and rllib/agents (#33242)

Signed-off-by: Artur Niederfahrenhorst <[email protected]>

rllib/algorithms/a2c/__init__.py

@@ -1,3 +1,3 @@
-from ray.rllib.algorithms.a2c.a2c import A2CConfig, A2C, A2C_DEFAULT_CONFIG
+from ray.rllib.algorithms.a2c.a2c import A2CConfig, A2C
 
-__all__ = ["A2CConfig", "A2C", "A2C_DEFAULT_CONFIG"]
+__all__ = ["A2CConfig", "A2C"]

rllib/algorithms/a2c/a2c.py

@@ -10,7 +10,6 @@
 )
 from ray.rllib.policy.sample_batch import DEFAULT_POLICY_ID
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 from ray.rllib.utils.metrics import (
     APPLY_GRADS_TIMER,
     COMPUTE_GRADS_TIMER,
@@ -241,20 +240,3 @@ def training_step(self) -> ResultDict:
         train_results = {DEFAULT_POLICY_ID: info}
 
         return train_results
-
-
-# Deprecated: Use ray.rllib.algorithms.a2c.A2CConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(A2CConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.agents.a3c.a2c.A2C_DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.a2c.a2c.A2CConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-A2C_DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/a3c/__init__.py

@@ -1,3 +1,3 @@
-from ray.rllib.algorithms.a3c.a3c import A3CConfig, A3C, DEFAULT_CONFIG
+from ray.rllib.algorithms.a3c.a3c import A3CConfig, A3C
 
-__all__ = ["A3CConfig", "A3C", "DEFAULT_CONFIG"]
+__all__ = ["A3CConfig", "A3C"]

rllib/algorithms/a3c/a3c.py

@@ -6,7 +6,6 @@
 from ray.rllib.evaluation.rollout_worker import RolloutWorker
 from ray.rllib.policy.policy import Policy
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 from ray.rllib.utils.metrics import (
     APPLY_GRADS_TIMER,
     GRAD_WAIT_TIMER,
@@ -251,20 +250,3 @@ def sample_and_compute_grads(worker: RolloutWorker) -> Dict[str, Any]:
             )
 
         return learner_info_builder.finalize()
-
-
-# Deprecated: Use ray.rllib.algorithms.a3c.A3CConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(A3CConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.agents.a3c.a3c.DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.a3c.a3c.A3CConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/algorithm_config.py

@@ -159,9 +159,9 @@ def from_dict(cls, config_dict: dict) -> "AlgorithmConfig":
         """Creates an AlgorithmConfig from a legacy python config dict.
 
         Examples:
-            >>> from ray.rllib.algorithms.ppo.ppo import DEFAULT_CONFIG, PPOConfig
-            >>> ppo_config = PPOConfig.from_dict(DEFAULT_CONFIG)
-            >>> ppo = ppo_config.build(env="Pendulum-v1")
+            >>> from ray.rllib.algorithms.ppo.ppo import PPOConfig # doctest: +SKIP
+            >>> ppo_config = PPOConfig.from_dict({...}) # doctest: +SKIP
+            >>> ppo = ppo_config.build(env="Pendulum-v1") # doctest: +SKIP
 
         Args:
             config_dict: The legacy formatted python config dict for some algorithm.

rllib/algorithms/alpha_star/__init__.py

@@ -1,11 +1,9 @@
 from ray.rllib.algorithms.alpha_star.alpha_star import (
     AlphaStar,
     AlphaStarConfig,
-    DEFAULT_CONFIG,
 )
 
 __all__ = [
     "AlphaStar",
     "AlphaStarConfig",
-    "DEFAULT_CONFIG",
 ]

rllib/algorithms/alpha_star/alpha_star.py

@@ -21,7 +21,6 @@
 from ray.rllib.policy.sample_batch import MultiAgentBatch
 from ray.rllib.utils import deep_update
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 from ray.rllib.utils.from_config import from_config
 from ray.rllib.utils.metrics import (
     LAST_TARGET_UPDATE_TS,
@@ -635,20 +634,3 @@ def __setstate__(self, state: dict) -> None:
         state_copy = state.copy()
         self.league_builder.__setstate__(state.pop("league_builder", {}))
         super().__setstate__(state_copy)
-
-
-# Deprecated: Use ray.rllib.algorithms.alpha_star.AlphaStarConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(AlphaStarConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.algorithms.alpha_star.alpha_star.DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.alpha_star.alpha_star.AlphaStarConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/alpha_zero/__init__.py

@@ -1,13 +1,11 @@
 from ray.rllib.algorithms.alpha_zero.alpha_zero import (
     AlphaZero,
     AlphaZeroConfig,
-    DEFAULT_CONFIG,
 )
 from ray.rllib.algorithms.alpha_zero.alpha_zero_policy import AlphaZeroPolicy
 
 __all__ = [
     "AlphaZero",
     "AlphaZeroConfig",
     "AlphaZeroPolicy",
-    "DEFAULT_CONFIG",
 ]

rllib/algorithms/alpha_zero/alpha_zero.py

@@ -16,7 +16,7 @@
 from ray.rllib.models.torch.torch_action_dist import TorchCategorical
 from ray.rllib.policy.policy import Policy
 from ray.rllib.policy.sample_batch import concat_samples
-from ray.rllib.utils.annotations import Deprecated, override
+from ray.rllib.utils.annotations import override
 from ray.rllib.utils.deprecation import DEPRECATED_VALUE
 from ray.rllib.utils.framework import try_import_torch
 from ray.rllib.utils.metrics import (
@@ -400,20 +400,3 @@ def training_step(self) -> ResultDict:
 
         # Return all collected metrics for the iteration.
         return train_results
-
-
-# Deprecated: Use ray.rllib.algorithms.alpha_zero.AlphaZeroConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(AlphaZeroConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.algorithms.alpha_zero.alpha_zero.DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.alpha_zero.alpha_zero.AlphaZeroConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/apex_ddpg/__init__.py

@@ -1,11 +1,9 @@
 from ray.rllib.algorithms.apex_ddpg.apex_ddpg import (
     ApexDDPG,
     ApexDDPGConfig,
-    APEX_DDPG_DEFAULT_CONFIG,
 )
 
 __all__ = [
     "ApexDDPG",
     "ApexDDPGConfig",
-    "APEX_DDPG_DEFAULT_CONFIG",
 ]

rllib/algorithms/apex_ddpg/apex_ddpg.py

@@ -4,7 +4,7 @@
 from ray.rllib.algorithms.apex_dqn.apex_dqn import ApexDQN
 from ray.rllib.algorithms.ddpg.ddpg import DDPG, DDPGConfig
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import DEPRECATED_VALUE, Deprecated
+from ray.rllib.utils.deprecation import DEPRECATED_VALUE
 from ray.rllib.utils.typing import (
     ResultDict,
 )
@@ -147,20 +147,3 @@ def setup(self, config: AlgorithmConfig):
     def training_step(self) -> ResultDict:
         """Use APEX-DQN's training iteration function."""
         return ApexDQN.training_step(self)
-
-
-# Deprecated: Use ray.rllib.algorithms.apex_ddpg.ApexDDPGConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(ApexDDPGConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.algorithms.ddpg.apex.APEX_DDPG_DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.apex_ddpg.apex_ddpg::ApexDDPGConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-APEX_DDPG_DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/apex_dqn/__init__.py

@@ -1,11 +1,9 @@
 from ray.rllib.algorithms.apex_dqn.apex_dqn import (
     ApexDQN,
     ApexDQNConfig,
-    APEX_DEFAULT_CONFIG,
 )
 
 __all__ = [
     "ApexDQN",
     "ApexDQNConfig",
-    "APEX_DEFAULT_CONFIG",
 ]

rllib/algorithms/apex_dqn/apex_dqn.py

@@ -29,7 +29,7 @@
 from ray.rllib.utils.actor_manager import FaultTolerantActorManager
 from ray.rllib.utils.actors import create_colocated_actors
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import DEPRECATED_VALUE, Deprecated
+from ray.rllib.utils.deprecation import DEPRECATED_VALUE
 from ray.rllib.utils.metrics import (
     LAST_TARGET_UPDATE_TS,
     NUM_AGENT_STEPS_SAMPLED,
@@ -753,20 +753,3 @@ def default_resource_request(
             ),
             strategy=cf.placement_strategy,
         )
-
-
-# Deprecated: Use ray.rllib.algorithms.apex_dqn.ApexDQNConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(ApexDQNConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.agents.dqn.apex.APEX_DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.apex_dqn.apex_dqn.ApexDQNConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-APEX_DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/appo/__init__.py

@@ -1,4 +1,4 @@
-from ray.rllib.algorithms.appo.appo import APPO, APPOConfig, DEFAULT_CONFIG
+from ray.rllib.algorithms.appo.appo import APPO, APPOConfig
 from ray.rllib.algorithms.appo.appo_tf_policy import APPOTF1Policy, APPOTF2Policy
 from ray.rllib.algorithms.appo.appo_torch_policy import APPOTorchPolicy
 
@@ -8,5 +8,4 @@
     "APPOTF1Policy",
     "APPOTF2Policy",
     "APPOTorchPolicy",
-    "DEFAULT_CONFIG",
 ]

rllib/algorithms/appo/appo.py

@@ -18,7 +18,6 @@
 from ray.rllib.execution.common import _get_shared_metrics, STEPS_SAMPLED_COUNTER
 from ray.rllib.policy.policy import Policy
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 from ray.rllib.utils.metrics import (
     LAST_TARGET_UPDATE_TS,
     NUM_AGENT_STEPS_SAMPLED,
@@ -294,20 +293,3 @@ def get_default_policy_class(
             from ray.rllib.algorithms.appo.appo_tf_policy import APPOTF2Policy
 
             return APPOTF2Policy
-
-
-# Deprecated: Use ray.rllib.algorithms.appo.APPOConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(APPOConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.agents.ppo.appo::DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.appo.appo::APPOConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/ars/__init__.py

@@ -1,4 +1,4 @@
-from ray.rllib.algorithms.ars.ars import ARS, ARSConfig, DEFAULT_CONFIG
+from ray.rllib.algorithms.ars.ars import ARS, ARSConfig
 from ray.rllib.algorithms.ars.ars_tf_policy import ARSTFPolicy
 from ray.rllib.algorithms.ars.ars_torch_policy import ARSTorchPolicy
 
@@ -7,5 +7,4 @@
     "ARSConfig",
     "ARSTFPolicy",
     "ARSTorchPolicy",
-    "DEFAULT_CONFIG",
 ]

rllib/algorithms/ars/ars.py

@@ -21,7 +21,6 @@
 from ray.rllib.utils import FilterManager
 from ray.rllib.utils.actor_manager import FaultAwareApply
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 from ray.rllib.utils.metrics import (
     NUM_AGENT_STEPS_SAMPLED,
     NUM_AGENT_STEPS_TRAINED,
@@ -605,20 +604,3 @@ def __setstate__(self, state):
         FilterManager.synchronize(
             {DEFAULT_POLICY_ID: self.policy.observation_filter}, self.workers
         )
-
-
-# Deprecated: Use ray.rllib.algorithms.ars.ARSConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(ARSConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.algorithms.ars.ars.DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.ars.ars.ARSConfig(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/ars/ars_torch_policy.py

@@ -13,7 +13,7 @@
     name="ARSTorchPolicy",
     framework="torch",
     loss_fn=None,
-    get_default_config=lambda: ray.rllib.algorithms.ars.ars.DEFAULT_CONFIG,
+    get_default_config=lambda: ray.rllib.algorithms.ars.ars.ARSConfig(),
     before_init=before_init,
     after_init=after_init,
     make_model_and_action_dist=make_model_and_action_dist,

rllib/algorithms/bandit/bandit.py

@@ -7,7 +7,6 @@
 from ray.rllib.algorithms.bandit.bandit_torch_policy import BanditTorchPolicy
 from ray.rllib.policy.policy import Policy
 from ray.rllib.utils.annotations import override
-from ray.rllib.utils.deprecation import Deprecated
 
 logger = logging.getLogger(__name__)
 
@@ -121,20 +120,3 @@ def get_default_policy_class(
             return BanditTFPolicy
         else:
             raise NotImplementedError("Only `framework=[torch|tf2]` supported!")
-
-
-# Deprecated: Use ray.rllib.algorithms.bandit.BanditLinUCBConfig instead!
-class _deprecated_default_config(dict):
-    def __init__(self):
-        super().__init__(BanditLinUCBConfig().to_dict())
-
-    @Deprecated(
-        old="ray.rllib.algorithms.bandit.bandit.DEFAULT_CONFIG",
-        new="ray.rllib.algorithms.bandit.bandit.BanditLin[UCB|TS]Config(...)",
-        error=True,
-    )
-    def __getitem__(self, item):
-        return super().__getitem__(item)
-
-
-DEFAULT_CONFIG = _deprecated_default_config()

rllib/algorithms/bandit/bandit_tf_policy.py

@@ -149,7 +149,7 @@ def after_init(policy, *args):
 
 BanditTFPolicy = build_tf_policy(
     name="BanditTFPolicy",
-    get_default_config=lambda: ray.rllib.algorithms.bandit.bandit.DEFAULT_CONFIG,
+    get_default_config=lambda: ray.rllib.algorithms.bandit.bandit.BanditConfig(),
     validate_spaces=validate_spaces,
     make_model=make_model,
     loss_fn=None,

rllib/algorithms/bc/__init__.py

@@ -1,8 +1,6 @@
-from ray.rllib.algorithms.bc.bc import BCConfig, BC, BC_DEFAULT_CONFIG
+from ray.rllib.algorithms.bc.bc import BCConfig, BC
 
 __all__ = [
     "BCConfig",
     "BC",
-    # Deprecated.
-    "BC_DEFAULT_CONFIG",
 ]