multi-agent-rlenv 3.3.0__py3-none-any.whl → 3.3.2__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__ = "3.3.0"
+__version__ = "3.3.2"
 
 from . import models
 from . import wrappers

marlenv/adapters/__init__.py

@@ -1,32 +1,42 @@
+from importlib.util import find_spec
 from .pymarl_adapter import PymarlAdapter
-from typing import Any
 
-__all__ = ["PymarlAdapter"]
-try:
+HAS_GYM = False
+if find_spec("gymnasium") is not None:
     from .gym_adapter import Gym
 
-    __all__.append("Gym")
-except ImportError:
-    Gym = Any
+    HAS_GYM = True
 
-try:
+HAS_PETTINGZOO = False
+if find_spec("pettingzoo") is not None:
     from .pettingzoo_adapter import PettingZoo
 
-    __all__.append("PettingZoo")
-except ImportError:
-    PettingZoo = Any
+    HAS_PETTINGZOO = True
 
-try:
+HAS_SMAC = False
+if find_spec("smac") is not None:
     from .smac_adapter import SMAC
 
-    __all__.append("SMAC")
-except ImportError:
-    SMAC = Any
+    HAS_SMAC = True
 
+HAS_OVERCOOKED = False
+if find_spec("overcooked_ai_py.mdp") is not None:
+    import numpy
 
-try:
+    # Overcooked assumes a version of numpy <2.0 where np.Inf is available.
+    setattr(numpy, "Inf", numpy.inf)
     from .overcooked_adapter import Overcooked
 
-    __all__.append("Overcooked")
-except ImportError:
-    Overcooked = Any
+    HAS_OVERCOOKED = True
+
+__all__ = [
+    "PymarlAdapter",
+    "Gym",
+    "PettingZoo",
+    "SMAC",
+    "Overcooked",
+    "HAS_GYM",
+    "HAS_PETTINGZOO",
+    "HAS_SMAC",
+    "HAS_OVERCOOKED",
+]

marlenv/adapters/overcooked_adapter.py

@@ -1,41 +1,72 @@
-import numpy as np
-import pygame
-import cv2
 import sys
-from marlenv.models import MARLEnv, State, Observation, Step, DiscreteActionSpace
+from dataclasses import dataclass
 from typing import Literal, Sequence
+from copy import deepcopy
+
+import cv2
+import numpy as np
 import numpy.typing as npt
-from overcooked_ai_py.mdp.overcooked_mdp import OvercookedGridworld, Action
+import pygame
+from marlenv.models import ContinuousSpace, DiscreteActionSpace, MARLEnv, Observation, State, Step
+
 from overcooked_ai_py.mdp.overcooked_env import OvercookedEnv
+from overcooked_ai_py.mdp.overcooked_mdp import Action, OvercookedGridworld, OvercookedState
 from overcooked_ai_py.visualization.state_visualizer import StateVisualizer
-from dataclasses import dataclass
 
 
 @dataclass
 class Overcooked(MARLEnv[Sequence[int] | npt.NDArray, DiscreteActionSpace]):
+    horizon: int
+
     def __init__(self, oenv: OvercookedEnv):
         self._oenv = oenv
         assert isinstance(oenv.mdp, OvercookedGridworld)
         self._mdp = oenv.mdp
         self.visualizer = StateVisualizer()
+        shape = tuple(int(s) for s in self._mdp.get_lossless_state_encoding_shape())
+        shape = (shape[2], shape[0], shape[1])
         super().__init__(
-            action_space=DiscreteActionSpace(n_agents=self._mdp.num_players, n_actions=Action.NUM_ACTIONS),
-            observation_shape=(1,),
-            state_shape=(1,),
+            action_space=DiscreteActionSpace(
+                n_agents=self._mdp.num_players,
+                n_actions=Action.NUM_ACTIONS,
+                action_names=[Action.ACTION_TO_CHAR[a] for a in Action.ALL_ACTIONS],
+            ),
+            observation_shape=shape,
+            extras_shape=(1,),
+            extras_meanings=["timestep"],
+            state_shape=shape,
+            state_extra_shape=(1,),
+            reward_space=ContinuousSpace.from_shape(1),
         )
+        self.horizon = int(self._oenv.horizon)
+
+    @property
+    def state(self) -> OvercookedState:
+        """Current state of the environment"""
+        return self._oenv.state
+
+    def set_state(self, state: State):
+        raise NotImplementedError("Not yet implemented")
+
+    @property
+    def time_step(self):
+        return self.state.timestep
 
     def _state_data(self):
-        state = self._oenv.state
-        state = np.array(self._mdp.lossless_state_encoding(state))
+        state = np.array(self._mdp.lossless_state_encoding(self.state))
         # Use axes (agents, channels, height, width) instead of (agents, height, width, channels)
         state = np.transpose(state, (0, 3, 1, 2))
         return state
 
     def get_state(self):
-        return State(self._state_data())
+        return State(self._state_data()[0], np.array([self.time_step / self.horizon]))
 
     def get_observation(self) -> Observation:
-        return Observation(self._state_data(), self.available_actions())
+        return Observation(
+            data=self._state_data(),
+            available_actions=self.available_actions(),
+            extras=np.array([[self.time_step / self.horizon]] * self.n_agents),
+        )
 
     def available_actions(self):
         available_actions = np.full((self.n_agents, self.n_actions), False)
@@ -57,6 +88,16 @@ class Overcooked(MARLEnv[Sequence[int] | npt.NDArray, DiscreteActionSpace]):
             info=info,
         )
 
+    def __deepcopy__(self, memo: dict):
+        mdp = deepcopy(self._mdp)
+        return Overcooked(OvercookedEnv.from_mdp(mdp, horizon=self.horizon))
+
+    def __getstate__(self):
+        return {"horizon": self.horizon, "mdp": self._mdp}
+
+    def __setstate__(self, state: dict):
+        self.__init__(OvercookedEnv.from_mdp(state["mdp"], horizon=state["horizon"]))
+
     def get_image(self):
         rewards_dict = {}  # dictionary of details you want rendered in the UI
         for key, value in self._oenv.game_stats.items():

marlenv/env_builder.py

@@ -1,32 +1,27 @@
 from dataclasses import dataclass
 from typing import Generic, Literal, Optional, TypeVar, overload
-
 import numpy as np
 import numpy.typing as npt
 
 from . import wrappers
+from marlenv import adapters
 from .models import ActionSpace, MARLEnv
-from .adapters import PettingZoo
 
 A = TypeVar("A")
 AS = TypeVar("AS", bound=ActionSpace)
 
-try:
+if adapters.HAS_PETTINGZOO:
+    from .adapters import PettingZoo
     from pettingzoo import ParallelEnv
 
     @overload
-    def make(
-        env: ParallelEnv,
-    ) -> PettingZoo: ...
-
-    HAS_PETTINGZOO = True
-except ImportError:
-    HAS_PETTINGZOO = False
+    def make(env: ParallelEnv) -> PettingZoo: ...
 
 
-try:
-    from gymnasium import Env
+if adapters.HAS_GYM:
     from .adapters import Gym
+    from gymnasium import Env
+    import gymnasium
 
     @overload
     def make(env: Env) -> Gym: ...
@@ -37,25 +32,21 @@ try:
         Make an RLEnv from the `gymnasium` registry (e.g: "CartPole-v1").
         """
 
-    HAS_GYM = True
-except ImportError:
-    HAS_GYM = False
 
-try:
-    from smac.env import StarCraft2Env
+if adapters.HAS_SMAC:
     from .adapters import SMAC
+    from smac.env import StarCraft2Env
 
     @overload
     def make(env: StarCraft2Env) -> SMAC: ...
 
-    HAS_SMAC = True
-except ImportError:
-    HAS_SMAC = False
 
+if adapters.HAS_OVERCOOKED:
+    from .adapters import Overcooked
+    from overcooked_ai_py.mdp.overcooked_env import OvercookedEnv
 
-@overload
-def make(env: MARLEnv[A, AS]) -> MARLEnv[A, AS]:
-    """Why would you do this ?"""
+    @overload
+    def make(env: OvercookedEnv) -> Overcooked: ...
 
 
 def make(env, **kwargs):
@@ -64,32 +55,18 @@ def make(env, **kwargs):
         case MARLEnv():
             return env
         case str(env_id):
-            try:
-                import gymnasium
-            except ImportError:
-                raise ImportError("Gymnasium is not installed !")
-            from marlenv.adapters import Gym
-
-            gym_env = gymnasium.make(env_id, render_mode="rgb_array", **kwargs)
-            return Gym(gym_env)
-
-    try:
-        from marlenv.adapters import PettingZoo
-
-        if isinstance(env, ParallelEnv):
-            return PettingZoo(env)
-    except ImportError:
-        pass
-    try:
-        from smac.env import StarCraft2Env
-
-        from marlenv.adapters import SMAC
-
-        if isinstance(env, StarCraft2Env):
-            return SMAC(env)
-    except ImportError:
-        pass
-
+            if adapters.HAS_GYM:
+                gym_env = gymnasium.make(env_id, render_mode="rgb_array", **kwargs)
+                return Gym(gym_env)
+
+    if adapters.HAS_PETTINGZOO and isinstance(env, ParallelEnv):
+        return PettingZoo(env)  # type: ignore
+    if adapters.HAS_SMAC and isinstance(env, StarCraft2Env):
+        return SMAC(env)
+    if adapters.HAS_OVERCOOKED and isinstance(env, OvercookedEnv):
+        return Overcooked(env)  # type: ignore
+    if adapters.HAS_GYM and isinstance(env, Env):
+        return Gym(env)
     raise ValueError(f"Unknown environment type: {type(env)}")
 
 

marlenv/env_pool.py

@@ -1,6 +1,5 @@
 from typing import Sequence
 from dataclasses import dataclass
-import numpy as np
 import numpy.typing as npt
 from typing_extensions import TypeVar
 import random

marlenv/models/observation.py

@@ -58,9 +58,14 @@ class Observation:
             available_actions=self.available_actions[agent_id],
         )
 
+    @property
+    def shape(self) -> tuple[int, ...]:
+        """The individual shape of the observation data"""
+        return self.data[0].shape
+
     @property
     def extras_shape(self) -> tuple[int, ...]:
-        """The shape of the observation extras"""
+        """The individual shape of the observation extras"""
         return self.extras[0].shape
 
     def __hash__(self):

{multi_agent_rlenv-3.3.0.dist-info → multi_agent_rlenv-3.3.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.3.0
+Version: 3.3.2
 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>
@@ -8,16 +8,58 @@ License-File: LICENSE
 Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
 Requires-Python: <4,>=3.10
-Requires-Dist: gymnasium>=0.29.1
 Requires-Dist: numpy>=2.0.0
-Requires-Dist: opencv-python>=4.10.0.84
+Requires-Dist: opencv-python>=4.0
+Requires-Dist: typing-extensions>=4.0
+Provides-Extra: all
+Requires-Dist: gymnasium>0.29.1; extra == 'all'
+Requires-Dist: overcooked-ai; extra == 'all'
+Requires-Dist: pettingzoo>=1.20; extra == 'all'
+Requires-Dist: pymunk>=6.0; extra == 'all'
+Requires-Dist: pysc2; extra == 'all'
+Requires-Dist: scipy>=1.10; extra == 'all'
+Requires-Dist: smac; extra == 'all'
+Provides-Extra: gym
+Requires-Dist: gymnasium>=0.29.1; extra == 'gym'
+Provides-Extra: overcooked
+Requires-Dist: overcooked-ai>=1.1.0; extra == 'overcooked'
+Requires-Dist: scipy>=1.10; extra == 'overcooked'
+Provides-Extra: pettingzoo
+Requires-Dist: pettingzoo>=1.20; extra == 'pettingzoo'
+Requires-Dist: pymunk>=6.0; extra == 'pettingzoo'
+Requires-Dist: scipy>=1.10; extra == 'pettingzoo'
+Provides-Extra: smac
+Requires-Dist: pysc2; extra == 'smac'
+Requires-Dist: smac; extra == 'smac'
 Description-Content-Type: text/markdown
 
-# `marlenv` - A unified interface for muti-agent reinforcement learning
+# `marlenv` - A unified framework for muti-agent reinforcement learning
+**Documentation: [https://yamoling.github.io/multi-agent-rlenv](https://yamoling.github.io/multi-agent-rlenv)**
+
 The objective of `marlenv` is to provide a common (typed) interface for many different reinforcement learning environments.
 
 As such, `marlenv` provides high level abstractions of RL concepts such as `Observation`s or `Transition`s that are commonly represented as mere (confusing) lists or tuples.
 
+## Installation
+Install with you preferred package manager (`uv`, `pip`, `poetry`, ...):
+```bash
+$ pip install marlenv[all] # Enable all features
+$ pip install marlenv      # Basic installation
+```
+
+There are multiple optional dependencies if you want to support specific libraries and environments. Available options are:
+- `smac` for StarCraft II environments
+- `gym` for OpenAI Gym environments
+- `pettingzoo` for PettingZoo environments
+- `overcooked` for Overcooked environments
+
+Install them with:
+```bash
+$ pip install marlenv[smac] # Install SMAC
+$ pip install marlenv[gym,smac]  # Install Gym & smac support
+```
+
+
 ## Using `marlenv` with existing libraries
 `marlenv` unifies multiple popular libraries under a single interface. Namely, `marlenv` supports `smac`, `gymnasium` and `pettingzoo`.
 
@@ -47,7 +89,7 @@ from marlenv import RLEnv, DiscreteActionSpace, Observation
 N_AGENTS = 3
 N_ACTIONS = 5
 
-class CustomEnv(RLEnv[DiscreteActionSpace]):
+class CustomEnv(MARLEnv[DiscreteActionSpace]):
     def __init__(self, width: int, height: int):
         super().__init__(
             action_space=DiscreteActionSpace(N_AGENTS, N_ACTIONS),

{multi_agent_rlenv-3.3.0.dist-info → multi_agent_rlenv-3.3.2.dist-info}/RECORD

@@ -1,19 +1,19 @@
-marlenv/__init__.py,sha256=G_8rNDcgzG_colnRNhbX04DIwSZWMJ2Eh7GSHqgIgzo,3741
-marlenv/env_builder.py,sha256=YH8yMV74t_joJmyUjBaTB1JabpdugFMY9R8jp2QdRdE,5700
-marlenv/env_pool.py,sha256=TSSYwD5-g4G473Ea097wFVbp3tyQrawywLIAFFEJCJY,1089
+marlenv/__init__.py,sha256=GFEcoE8jkA0vbuQyrVRpgsS-iLIelxjpwsB_6pGJGjs,3741
+marlenv/env_builder.py,sha256=_rdwcWRqnHP7i4M4Oje1Y2nrEBKH9EzTpqOuw_PNUyw,5560
+marlenv/env_pool.py,sha256=R3WIrnQ5Zvff4HR1ecfkDmuO2zl7v1ywQ0K2_nvWFzs,1070
 marlenv/exceptions.py,sha256=gJUC_2rVAvOfK_ypVFc7Myh-pIfSU3To38VBVS_0rZA,1179
 marlenv/mock_env.py,sha256=qB0fYFIfbopJf7Va8kCeVI5vsOy1-2JdEYe9gdV1Ruw,4761
 marlenv/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-marlenv/adapters/__init__.py,sha256=LMUebMfyj0Or2luiTlgeSqF2uKjp78Ez8hPDjVUrgVQ,554
+marlenv/adapters/__init__.py,sha256=NEmuHPWz4SGQcgF7QuIeA0QaXK141JoYco-7mqj9Ghk,883
 marlenv/adapters/gym_adapter.py,sha256=Vx6ZrYI7kiNlJODmqyjXu9WCdbCr6trcMNot0pvYD74,2864
-marlenv/adapters/overcooked_adapter.py,sha256=IR2hVMb-ZUxl3BP8WELJGJJBWhSVbuJYjEgg1aBHj20,4774
+marlenv/adapters/overcooked_adapter.py,sha256=Yf1xmjSgDOOSzR3QNqjxQ_ROFEdKUe_C8BF2nLlW3Us,6162
 marlenv/adapters/pettingzoo_adapter.py,sha256=9rwSc_b7qV3ChtEIevOkJvtIp7WoY3CVnu6L9DxlMB4,2852
 marlenv/adapters/pymarl_adapter.py,sha256=x__E90XpFbfSWhnBHtkcD6WYkmKki1LByNbUFoDBUcg,3416
 marlenv/adapters/smac_adapter.py,sha256=fOfKo1hL4ioKtM5qQGcwtfdkdwUEACjAZqaGmkoQUcU,8373
 marlenv/models/__init__.py,sha256=9M-rnj94nsdyO4zm_VEtyYBmde3iD2_eIY4bMB-IqCo,555
 marlenv/models/env.py,sha256=faezAKOIccBauOFeo9wu5sX32pFmP3AMmGyJzaTRJcM,7514
 marlenv/models/episode.py,sha256=ZGBx6lb2snrUhDgFEwHPV1dp-XvMA7k4quQVUNQxsP0,15140
-marlenv/models/observation.py,sha256=rTAesS_jaIyRlH4wjo2izEpWS0Hn5_UKjhbvdp0H4tA,2994
+marlenv/models/observation.py,sha256=kAmh1hIoC2TGrZlGVzV0y4TXXCSrI7gcmG0raeoncYk,3153
 marlenv/models/spaces.py,sha256=pw8Sum_fHBkR-lyfTqUij4azMCNm8oBZrYZe4WVR7rA,7652
 marlenv/models/state.py,sha256=958PXTHadi3gtRnhGgcGtqBnF44R11kdcx62NN2gwxA,1717
 marlenv/models/step.py,sha256=LKGAV2Cu-k9Gz1hwrfvGx51l8axtQRqDE9WVL5r2A1Q,3037
@@ -31,7 +31,7 @@ marlenv/wrappers/penalty_wrapper.py,sha256=v4_H8OEN2-yujLzRb6P7W7KwmXHtjAFsxcdp3
 marlenv/wrappers/rlenv_wrapper.py,sha256=C2XekgBIM4x3Wa2Mtsn7rihRD4ymC2hORI473Af0sfw,2962
 marlenv/wrappers/time_limit.py,sha256=CDIMMJPMyIDHSFxUJaC7nb7Kd86-07NgZeFhrpZm82o,3985
 marlenv/wrappers/video_recorder.py,sha256=d5AFu6qHqby9mOcBsYWYPxAPiK1vtnfMYdZ81AnCekI,2624
-multi_agent_rlenv-3.3.0.dist-info/METADATA,sha256=V7uoHPEZbXppxZHkO63e1UOm8AwkQ1x5jeF16BiNoIQ,3357
-multi_agent_rlenv-3.3.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-multi_agent_rlenv-3.3.0.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
-multi_agent_rlenv-3.3.0.dist-info/RECORD,,
+multi_agent_rlenv-3.3.2.dist-info/METADATA,sha256=s9O2h4QdJZ4Ytq1hRiBFJPsAtWV0bd1JLxDx4MChaI0,4897
+multi_agent_rlenv-3.3.2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+multi_agent_rlenv-3.3.2.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
+multi_agent_rlenv-3.3.2.dist-info/RECORD,,