multi-agent-rlenv 3.5.1__py3-none-any.whl → 3.5.4__py3-none-any.whl

marlenv/__init__.py

@@ -62,7 +62,7 @@ print(env.extras_shape) # (1, )
 If you want to create a new environment, you can simply create a class that inherits from `MARLEnv`. If you want to create a wrapper around an existing `MARLEnv`, you probably want to subclass `RLEnvWrapper` which implements a default behaviour for every method.
 """
 
-__version__ = "3.5.1"
+__version__ = "3.5.4"
 
 from . import models
 from .models import (

marlenv/models/env.py

@@ -199,7 +199,7 @@ class MARLEnv(ABC, Generic[ActionSpaceType]):
             episode.add(step, action)
         return episode
 
-    def has_same_inouts(self, other) -> bool:
+    def has_same_inouts(self, other: "MARLEnv[ActionSpaceType]") -> bool:
         """Alias for `have_same_inouts(self, other)`."""
         if not isinstance(other, MARLEnv):
             return False

marlenv/models/episode.py

@@ -22,10 +22,10 @@ class Episode:
     all_extras: list[npt.NDArray[np.float32]]
     actions: list[npt.NDArray]
     rewards: list[npt.NDArray[np.float32]]
-    all_available_actions: list[npt.NDArray[np.bool_]]
+    all_available_actions: list[npt.NDArray[np.bool]]
     all_states: list[npt.NDArray[np.float32]]
     all_states_extras: list[npt.NDArray[np.float32]]
-    metrics: dict[str, float]
+    metrics: dict[str, Any]
     episode_len: int
     other: dict[str, list[Any]]
     is_done: bool = False
@@ -33,7 +33,7 @@ class Episode:
     """Whether the episode did reach a terminal state (different from truncated)"""
 
     @staticmethod
-    def new(obs: Observation, state: State, metrics: Optional[dict[str, float]] = None) -> "Episode":
+    def new(obs: Observation, state: State, metrics: Optional[dict[str, Any]] = None) -> "Episode":
         if metrics is None:
             metrics = {}
         return Episode(
@@ -66,13 +66,13 @@ class Episode:
         if target_len < self.episode_len:
             raise ValueError(f"Cannot pad episode to a smaller size: {target_len} < {self.episode_len}")
         padding_size = target_len - self.episode_len
-        obs = self.all_observations + [self.all_observations[0]] * padding_size
-        extras = self.all_extras + [self.all_extras[0]] * padding_size
-        actions = self.actions + [self.actions[0]] * padding_size
-        rewards = self.rewards + [self.rewards[0]] * padding_size
+        obs = self.all_observations + [np.zeros_like(self.all_observations[0])] * padding_size
+        extras = self.all_extras + [np.zeros_like(self.all_extras[0])] * padding_size
+        actions = self.actions + [np.zeros_like(self.actions[0])] * padding_size
+        rewards = self.rewards + [np.zeros_like(self.rewards[0])] * padding_size
         availables = self.all_available_actions + [self.all_available_actions[0]] * padding_size
-        states = self.all_states + [self.all_states[0]] * padding_size
-        states_extras = self.all_states_extras + [self.all_states_extras[0]] * padding_size
+        states = self.all_states + [np.zeros_like(self.all_states[0])] * padding_size
+        states_extras = self.all_states_extras + [np.zeros_like(self.all_states_extras[0])] * padding_size
         other = {key: value + [value[0]] * padding_size for key, value in self.other.items()}
         return Episode(
             all_observations=obs,
@@ -363,51 +363,6 @@ class Episode:
             for i, s in enumerate(scores):
                 self.metrics[f"score-{i}"] = float(s)
 
-    # def add_data(
-    #     self,
-    #     new_obs: Observation,
-    #     new_state: State,
-    #     action: A,
-    #     reward: np.ndarray,
-    #     done: bool,
-    #     truncated: bool,
-    #     info: dict[str, Any],
-    #     **kwargs,
-    # ):
-    #     """Add a new transition to the episode"""
-    #     self.episode_len += 1
-    #     self.all_observations.append(new_obs.data)
-    #     self.all_extras.append(new_obs.extras)
-    #     self.all_available_actions.append(new_obs.available_actions)
-    #     self.all_states.append(new_state.data)
-    #     self.all_states_extras.append(new_state.extras)
-    #     match action:
-    #         case np.ndarray() as action:
-    #             self.actions.append(action)
-    #         case other:
-    #             self.actions.append(np.array(other))
-    #     self.rewards.append(reward)
-    #     for key, value in kwargs.items():
-    #         current = self.other.get(key, [])
-    #         current.append(value)
-    #         self.other[key] = current
-
-    #     if done:
-    #         # Only set the truncated flag if the episode is not done (both could happen with a time limit)
-    #         self.is_truncated = truncated
-    #         self.is_done = done
-    #         # Add metrics that can be plotted
-    #         for key, value in info.items():
-    #             if isinstance(value, bool):
-    #                 value = int(value)
-    #             self.metrics[key] = value
-    #         self.metrics["episode_len"] = self.episode_len
-
-    #         rewards = np.array(self.rewards)
-    #         scores = np.sum(rewards, axis=0)
-    #         for i, s in enumerate(scores):
-    #             self.metrics[f"score-{i}"] = float(s)
-
-    def add_metrics(self, metrics: dict[str, float]):
+    def add_metrics(self, metrics: dict[str, Any]):
         """Add metrics to the episode"""
         self.metrics.update(metrics)

marlenv/models/observation.py

@@ -87,3 +87,13 @@ class Observation:
             if not np.array_equal(self.data, other.data):
                 return False
         return np.array_equal(self.extras, other.extras) and np.array_equal(self.available_actions, other.available_actions)
+
+    def as_tensors(self, device=None):
+        """
+        Convert the observation to a tuple of tensors of shape (1, n_agents, <dim>).
+        """
+        import torch
+
+        data = torch.from_numpy(self.data).unsqueeze(0).to(device, non_blocking=True)
+        extras = torch.from_numpy(self.extras).unsqueeze(0).to(device, non_blocking=True)
+        return data, extras

marlenv/models/state.py

@@ -52,3 +52,11 @@ class State(Generic[StateType]):
         if not np.array_equal(self.extras, value.extras):
             return False
         return True
+
+    def as_tensors(self, device=None):
+        """Convert the state to a tuple of tensors of shape (1, <dim>)."""
+        import torch
+
+        data = torch.from_numpy(self.data).unsqueeze(0).to(device, non_blocking=True)
+        extras = torch.from_numpy(self.extras).unsqueeze(0).to(device, non_blocking=True)
+        return data, extras

marlenv/wrappers/potential_shaping.py

@@ -2,10 +2,15 @@ from abc import abstractmethod, ABC
 from .rlenv_wrapper import RLEnvWrapper
 from marlenv import Space, MARLEnv, Observation
 from typing import TypeVar, Optional
+import numpy as np
+import numpy.typing as npt
+
+from dataclasses import dataclass
 
 A = TypeVar("A", bound=Space)
 
 
+@dataclass
 class PotentialShaping(RLEnvWrapper[A], ABC):
     """
     Potential shaping for the Laser Learning Environment (LLE).
@@ -23,7 +28,7 @@ class PotentialShaping(RLEnvWrapper[A], ABC):
     ):
         super().__init__(env, extra_shape=extra_shape)
         self.gamma = gamma
-        self.current_potential = self.compute_potential()
+        self._current_potential = self.compute_potential()
 
     def add_extras(self, obs: Observation) -> Observation:
         """Add the extras related to potential shaping. Does nothing by default."""
@@ -31,19 +36,19 @@ class PotentialShaping(RLEnvWrapper[A], ABC):
 
     def reset(self):
         obs, state = super().reset()
-        self.current_potential = self.compute_potential()
+        self._current_potential = self.compute_potential()
         return self.add_extras(obs), state
 
     def step(self, actions):
-        phi_t = self.current_potential
+        prev_potential = self._current_potential
         step = super().step(actions)
 
-        self.current_potential = self.compute_potential()
-        shaped_reward = self.gamma * self.current_potential - phi_t
+        self._current_potential = self.compute_potential()
+        shaped_reward = self.gamma * self._current_potential - prev_potential
         step.obs = self.add_extras(step.obs)
         step.reward += shaped_reward
         return step
 
     @abstractmethod
-    def compute_potential(self) -> float:
+    def compute_potential(self) -> float | npt.NDArray[np.float32]:
         """Compute the potential of the current state of the environment."""

{multi_agent_rlenv-3.5.1.dist-info → multi_agent_rlenv-3.5.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.5.1
+Version: 3.5.4
 Summary: A strongly typed Multi-Agent Reinforcement Learning framework
 Project-URL: repository, https://github.com/yamoling/multi-agent-rlenv
 Author-email: Yannick Molinghen <yannick.molinghen@ulb.be>
@@ -19,6 +19,7 @@ Requires-Dist: pymunk>=6.0; extra == 'all'
 Requires-Dist: pysc2; extra == 'all'
 Requires-Dist: scipy>=1.10; extra == 'all'
 Requires-Dist: smac; extra == 'all'
+Requires-Dist: torch>=2.0; extra == 'all'
 Provides-Extra: gym
 Requires-Dist: gymnasium>=0.29.1; extra == 'gym'
 Provides-Extra: overcooked
@@ -31,6 +32,8 @@ Requires-Dist: scipy>=1.10; extra == 'pettingzoo'
 Provides-Extra: smac
 Requires-Dist: pysc2; extra == 'smac'
 Requires-Dist: smac; extra == 'smac'
+Provides-Extra: torch
+Requires-Dist: torch>=2.0; extra == 'torch'
 Description-Content-Type: text/markdown
 
 # `marlenv` - A unified framework for muti-agent reinforcement learning

{multi_agent_rlenv-3.5.1.dist-info → multi_agent_rlenv-3.5.4.dist-info}/RECORD

@@ -1,4 +1,4 @@
-marlenv/__init__.py,sha256=zNSSEveqBncupuGlVpeywLpK4BZSrXufi-BUhG96e4s,3656
+marlenv/__init__.py,sha256=-XMj91_t4PoY6zLhr395u_svfIXYlvk-tMR4YqG-2VQ,3656
 marlenv/env_builder.py,sha256=RJoHJLYAUE1ausAoJiRC3fUxyxpH1WRJf7Sdm2ml-uk,5517
 marlenv/env_pool.py,sha256=nCEBkGQU62fcvCAANyAqY8gCFjYlVnSCg-V3Fhx00yc,933
 marlenv/exceptions.py,sha256=gJUC_2rVAvOfK_ypVFc7Myh-pIfSU3To38VBVS_0rZA,1179
@@ -11,11 +11,11 @@ marlenv/adapters/pettingzoo_adapter.py,sha256=w9Ta-X4L_6ZXdDGmREOdcU0vpLR8lGP__s
 marlenv/adapters/pymarl_adapter.py,sha256=2s7EY31s1hrml3q-BBaXo_eDMXTjkebozZPvzsgrb9c,3353
 marlenv/adapters/smac_adapter.py,sha256=8uWC7YKsaSXeTS8AUhpGOKvrWMbVEQT2-pml5BaFUB0,8343
 marlenv/models/__init__.py,sha256=uihmRs71Gg5z7Bvau_xtaQVg7xEtX8sTzi74bIHL5P0,443
-marlenv/models/env.py,sha256=KB3-LcvIbGG-78L8ZavfjKykoO85FvhZjs5lJQKvxE0,7813
-marlenv/models/episode.py,sha256=IF3-8YV0tHsIjTYZUOlHmX_IyjnrrzTWk-HPk_mwcR4,15100
-marlenv/models/observation.py,sha256=kAmh1hIoC2TGrZlGVzV0y4TXXCSrI7gcmG0raeoncYk,3153
+marlenv/models/env.py,sha256=BG1iVHxGD_p827mF0ewyOBn6wU2gtFsHLW1b4UtW-V0,7841
+marlenv/models/episode.py,sha256=DOX2FpWK-wm0BX_vC2bVD4BUMiP-JHUfRl80MGXD7kc,13472
+marlenv/models/observation.py,sha256=RhvKvmys4bu3UwwVsvu7fJ7TMKt2QkKnBD1e0hw2r7s,3528
 marlenv/models/spaces.py,sha256=v7jnhPfj7vq7DFFJpSbQEYe4NGLLlj_bj2pzvvSBX7Y,7777
-marlenv/models/state.py,sha256=958PXTHadi3gtRnhGgcGtqBnF44R11kdcx62NN2gwxA,1717
+marlenv/models/state.py,sha256=LbP--JxBzRwMFpEAaZyxCX13xKQ27xPE2fabohaq9YI,2058
 marlenv/models/step.py,sha256=00PhD_ccdCIYAY1SVJdJU91weU0Y_tNIJwK16TN_53I,3056
 marlenv/models/transition.py,sha256=UkJVRNxZoyRkjE7YmKtUf_4xA7cOEh20O60dTldbvys,5070
 marlenv/utils/__init__.py,sha256=C3qhvkVwctBP8mG3G5nkAZ5DKfErVRkdbHo7oeWVsM0,209
@@ -30,11 +30,11 @@ marlenv/wrappers/delayed_rewards.py,sha256=P8az9rYmu67OzL1ZEFqfTQcCxRI_AXKXrKUBQ
 marlenv/wrappers/last_action_wrapper.py,sha256=QVepSLcWExqACwKvAM0G2LALapSoWdd7YHmah2LZ3vE,2603
 marlenv/wrappers/paddings.py,sha256=0aAi7RP1yL8I5mR4Oxzl9-itKys88mgsPjqe7q-frbk,2024
 marlenv/wrappers/penalty_wrapper.py,sha256=3YBoUV6ETksZ8tFEOq1WYXvPs3ejMAehE6-QA8e4JOE,864
-marlenv/wrappers/potential_shaping.py,sha256=N4h0agFVxZZgCwMZTPFW64MGRFEYYFFVd9k4F27_MgQ,1520
+marlenv/wrappers/potential_shaping.py,sha256=T_QvnmWReCgpyoInxRw2UXbmdvcBD5U-vV1ledLG7y8,1661
 marlenv/wrappers/rlenv_wrapper.py,sha256=S6G1VjFklTEzU6bj0AXrTDXnsTQJARq8VB4uUH6AXe4,2993
 marlenv/wrappers/time_limit.py,sha256=GxbxcbfFyuVg14ylQU2C_cjmV9q4uDAt5wepfgX_PyM,3976
 marlenv/wrappers/video_recorder.py,sha256=ucBQSNRPqDr-2mYxrTCqlrWcxSWtSJ7XlRC9-LdukBM,2535
-multi_agent_rlenv-3.5.1.dist-info/METADATA,sha256=92aLf2gsXK8leco_rbbZzU-fLnIIgV3F6xdV1D3fP50,4897
-multi_agent_rlenv-3.5.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-multi_agent_rlenv-3.5.1.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
-multi_agent_rlenv-3.5.1.dist-info/RECORD,,
+multi_agent_rlenv-3.5.4.dist-info/METADATA,sha256=xgYpo8yrpncxWufHZSqMIeeT3Bb8Ll9DylTRKyesdT4,5005
+multi_agent_rlenv-3.5.4.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+multi_agent_rlenv-3.5.4.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
+multi_agent_rlenv-3.5.4.dist-info/RECORD,,