[RLlib] Introduce new OldAPIStack decorator; Do-over of all API decorators

rllib/connectors/action/clip.py

@@ -7,10 +7,10 @@
 from ray.rllib.connectors.registry import register_connector
 from ray.rllib.utils.spaces.space_utils import clip_action, get_base_struct_from_space
 from ray.rllib.utils.typing import ActionConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ClipActionsConnector(ActionConnector):
     def __init__(self, ctx: ConnectorContext):
         super().__init__(ctx)

rllib/connectors/action/immutable.py

@@ -9,10 +9,10 @@
 from ray.rllib.connectors.registry import register_connector
 from ray.rllib.utils.numpy import make_action_immutable
 from ray.rllib.utils.typing import ActionConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ImmutableActionsConnector(ActionConnector):
     def transform(self, ac_data: ActionConnectorDataType) -> ActionConnectorDataType:
         assert isinstance(

rllib/connectors/action/lambdas.py

@@ -12,10 +12,10 @@
     StateBatches,
     TensorStructType,
 )
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def register_lambda_action_connector(
     name: str, fn: Callable[[TensorStructType, StateBatches, Dict], PolicyOutputType]
 ) -> Type[ActionConnector]:
@@ -64,7 +64,7 @@ def from_state(ctx: ConnectorContext, params: Any):
 
 
 # Convert actions and states into numpy arrays if necessary.
-ConvertToNumpyConnector = PublicAPI(stability="alpha")(
+ConvertToNumpyConnector = OldAPIStack(
     register_lambda_action_connector(
         "ConvertToNumpyConnector",
         lambda actions, states, fetches: (

rllib/connectors/action/normalize.py

@@ -10,10 +10,10 @@
     unsquash_action,
 )
 from ray.rllib.utils.typing import ActionConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class NormalizeActionsConnector(ActionConnector):
     def __init__(self, ctx: ConnectorContext):
         super().__init__(ctx)

rllib/connectors/action/pipeline.py

@@ -9,15 +9,15 @@
     ConnectorPipeline,
 )
 from ray.rllib.connectors.registry import get_connector, register_connector
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.utils.typing import ActionConnectorDataType
-from ray.util.annotations import PublicAPI
 from ray.util.timer import _Timer
 
 
 logger = logging.getLogger(__name__)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ActionConnectorPipeline(ConnectorPipeline, ActionConnector):
     def __init__(self, ctx: ConnectorContext, connectors: List[Connector]):
         super().__init__(ctx, connectors)

rllib/connectors/agent/clip_reward.py

@@ -9,10 +9,10 @@
 from ray.rllib.connectors.registry import register_connector
 from ray.rllib.policy.sample_batch import SampleBatch
 from ray.rllib.utils.typing import AgentConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ClipRewardAgentConnector(AgentConnector):
     def __init__(self, ctx: ConnectorContext, sign=False, limit=None):
         super().__init__(ctx)

rllib/connectors/agent/env_sampling.py

@@ -6,10 +6,10 @@
 )
 from ray.rllib.connectors.registry import register_connector
 from ray.rllib.utils.typing import AgentConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class EnvSamplingAgentConnector(AgentConnector):
     def __init__(self, ctx: ConnectorContext, sign=False, limit=None):
         super().__init__(ctx)

rllib/connectors/agent/lambdas.py

@@ -13,10 +13,10 @@
     AgentConnectorDataType,
     AgentConnectorsOutput,
 )
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def register_lambda_agent_connector(
     name: str, fn: Callable[[Any], Any]
 ) -> Type[AgentConnector]:
@@ -54,7 +54,7 @@ def from_state(ctx: ConnectorContext, params: Any):
     return LambdaAgentConnector
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def flatten_data(data: AgentConnectorsOutput):
     assert isinstance(
         data, AgentConnectorsOutput
@@ -81,6 +81,6 @@ def flatten_data(data: AgentConnectorsOutput):
 
 # Agent connector to build and return a flattened observation SampleBatch
 # in addition to the original input dict.
-FlattenDataAgentConnector = PublicAPI(stability="alpha")(
+FlattenDataAgentConnector = OldAPIStack(
     register_lambda_agent_connector("FlattenDataAgentConnector", flatten_data)
 )

rllib/connectors/agent/mean_std_filter.py

@@ -11,15 +11,15 @@
 )
 from ray.rllib.connectors.registry import register_connector
 from ray.rllib.policy.sample_batch import SampleBatch
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.utils.filter import Filter
 from ray.rllib.utils.filter import MeanStdFilter, ConcurrentMeanStdFilter
 from ray.rllib.utils.spaces.space_utils import get_base_struct_from_space
 from ray.rllib.utils.typing import AgentConnectorDataType
-from ray.util.annotations import PublicAPI
 from ray.rllib.utils.filter import RunningStat
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class MeanStdObservationFilterAgentConnector(SyncedFilterAgentConnector):
     """A connector used to mean-std-filter observations.
 
@@ -149,7 +149,7 @@ def sync(self, other: "AgentConnector") -> None:
         return self.filter.sync(other.filter)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ConcurrentMeanStdObservationFilterAgentConnector(
     MeanStdObservationFilterAgentConnector
 ):

rllib/connectors/agent/obs_preproc.py

@@ -8,13 +8,10 @@
 from ray.rllib.models.preprocessors import get_preprocessor, NoPreprocessor
 from ray.rllib.policy.sample_batch import SampleBatch
 from ray.rllib.utils.typing import AgentConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
-# Bridging between current obs preprocessors and connector.
-# We should not introduce any new preprocessors.
-# TODO(jungong) : migrate and implement preprocessor library in Connector framework.
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ObsPreprocessorConnector(AgentConnector):
     """A connector that wraps around existing RLlib observation preprocessors.
 

rllib/connectors/agent/pipeline.py

@@ -10,14 +10,14 @@
 )
 from ray.rllib.connectors.registry import get_connector, register_connector
 from ray.rllib.utils.typing import ActionConnectorDataType, AgentConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.util.timer import _Timer
 
 
 logger = logging.getLogger(__name__)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class AgentConnectorPipeline(ConnectorPipeline, AgentConnector):
     def __init__(self, ctx: ConnectorContext, connectors: List[Connector]):
         super().__init__(ctx, connectors)

rllib/connectors/agent/state_buffer.py

@@ -18,13 +18,13 @@
 from ray.rllib.core.models.base import STATE_OUT
 from ray.rllib.utils.spaces.space_utils import get_base_struct_from_space
 from ray.rllib.utils.typing import ActionConnectorDataType, AgentConnectorDataType
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 
 logger = logging.getLogger(__name__)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class StateBufferConnector(AgentConnector):
     def __init__(self, ctx: ConnectorContext, states: Any = None):
         super().__init__(ctx)

rllib/connectors/agent/synced_filter.py

@@ -2,11 +2,11 @@
     AgentConnector,
     ConnectorContext,
 )
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.utils.filter import Filter
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class SyncedFilterAgentConnector(AgentConnector):
     """An agent connector that filters with synchronized parameters."""
 

rllib/connectors/agent/view_requirement.py

@@ -11,11 +11,11 @@
     AgentConnectorDataType,
     AgentConnectorsOutput,
 )
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.evaluation.collectors.agent_collector import AgentCollector
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ViewRequirementAgentConnector(AgentConnector):
     """This connector does 2 things:
     1. It filters data columns based on view_requirements for training and inference.

rllib/connectors/connector.py

@@ -14,15 +14,15 @@
     AlgorithmConfigDict,
     TensorType,
 )
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 
 if TYPE_CHECKING:
     from ray.rllib.policy.policy import Policy
 
 logger = logging.getLogger(__name__)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ConnectorContext:
     """Data bits that may be needed for running connectors.
 
@@ -79,7 +79,7 @@ def from_policy(policy: "Policy") -> "ConnectorContext":
         )
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class Connector(abc.ABC):
     """Connector base class.
 
@@ -137,7 +137,7 @@ def from_state(self, ctx: ConnectorContext, params: Any) -> "Connector":
         return NotImplementedError
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class AgentConnector(Connector):
     """Connector connecting user environments to RLlib policies.
 
@@ -277,7 +277,7 @@ def transform(self, ac_data: AgentConnectorDataType) -> AgentConnectorDataType:
         raise NotImplementedError
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ActionConnector(Connector):
     """Action connector connects policy outputs including actions,
     to user environments.
@@ -332,7 +332,7 @@ def transform(self, ac_data: ActionConnectorDataType) -> ActionConnectorDataType
         raise NotImplementedError
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 class ConnectorPipeline(abc.ABC):
     """Utility class for quick manipulation of a connector pipeline."""
 

rllib/connectors/registry.py

@@ -1,14 +1,14 @@
 """Registry of connector names for global access."""
 from typing import Any
 
-from ray.util.annotations import PublicAPI
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.connectors.connector import Connector, ConnectorContext
 
 
 ALL_CONNECTORS = dict()
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def register_connector(name: str, cls: Connector):
     """Register a connector for use with RLlib.
 
@@ -28,7 +28,7 @@ def register_connector(name: str, cls: Connector):
     ALL_CONNECTORS[name] = cls
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def get_connector(name: str, ctx: ConnectorContext, params: Any = None) -> Connector:
     # TODO(jungong) : switch the order of parameters man!!
     """Get a connector by its name and serialized config.

rllib/connectors/util.py

@@ -17,7 +17,7 @@
     MeanStdObservationFilterAgentConnector,
     ConcurrentMeanStdObservationFilterAgentConnector,
 )
-from ray.util.annotations import PublicAPI, DeveloperAPI
+from ray.rllib.utils.annotations import OldAPIStack
 from ray.rllib.connectors.agent.synced_filter import SyncedFilterAgentConnector
 
 if TYPE_CHECKING:
@@ -44,7 +44,7 @@ def __clip_rewards(config: "AlgorithmConfig"):
     return config.clip_rewards or config.is_atari
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def get_agent_connectors_from_config(
     ctx: ConnectorContext,
     config: "AlgorithmConfig",
@@ -78,7 +78,7 @@ def get_agent_connectors_from_config(
     return AgentConnectorPipeline(ctx, connectors)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def get_action_connectors_from_config(
     ctx: ConnectorContext,
     config: "AlgorithmConfig",
@@ -98,7 +98,7 @@ def get_action_connectors_from_config(
     return ActionConnectorPipeline(ctx, connectors)
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def create_connectors_for_policy(policy: "Policy", config: "AlgorithmConfig"):
     """Util to create agent and action connectors for a Policy.
 
@@ -120,7 +120,7 @@ def create_connectors_for_policy(policy: "Policy", config: "AlgorithmConfig"):
     logger.info(policy.action_connectors.__str__(indentation=4))
 
 
-@PublicAPI(stability="alpha")
+@OldAPIStack
 def restore_connectors_for_policy(
     policy: "Policy", connector_config: Tuple[str, Tuple[Any]]
 ) -> Connector:
@@ -136,7 +136,7 @@ def restore_connectors_for_policy(
 
 
 # We need this filter selection mechanism temporarily to remain compatible to old API
-@DeveloperAPI
+@OldAPIStack
 def get_synced_filter_connector(ctx: ConnectorContext):
     filter_specifier = ctx.config.get("observation_filter")
     if filter_specifier == "MeanStdFilter":
@@ -149,7 +149,7 @@ def get_synced_filter_connector(ctx: ConnectorContext):
         raise Exception("Unknown observation_filter: " + str(filter_specifier))
 
 
-@DeveloperAPI
+@OldAPIStack
 def maybe_get_filters_for_syncing(rollout_worker, policy_id):
     # As long as the historic filter synchronization mechanism is in
     # place, we need to put filters into self.filters so that they get

rllib/env/apis/task_settable_env.py

@@ -1,9 +1,12 @@
 import gymnasium as gym
 from typing import List, Any
 
+from ray.rllib.utils.annotations import OldAPIStack
+
 TaskType = Any  # Can be different types depending on env, e.g., int or dict
 
 
+@OldAPIStack
 class TaskSettableEnv(gym.Env):
     """
     Extension of gym.Env to define a task-settable Env.