multi-agent-rlenv 3.5.2__py3-none-any.whl → 3.5.5__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__ = "3.5.2"
+__version__ = "3.5.5"
 
 from . import models
 from .models import (

marlenv/models/episode.py

@@ -2,30 +2,32 @@ from dataclasses import dataclass
 from functools import cached_property
 from typing import Any, Callable, Optional, Sequence, overload
 
+import cv2
 import numpy as np
 import numpy.typing as npt
-import cv2
 
+from marlenv.exceptions import EnvironmentMismatchException, ReplayMismatchException
+from marlenv.utils import CachedPropertyInvalidator
+
+from .env import MARLEnv
 from .observation import Observation
 from .state import State
 from .step import Step
 from .transition import Transition
-from .env import MARLEnv
-from marlenv.exceptions import EnvironmentMismatchException, ReplayMismatchException
 
 
 @dataclass
-class Episode:
+class Episode(CachedPropertyInvalidator):
     """Episode model made of observations, actions, rewards, ..."""
 
     all_observations: list[npt.NDArray[np.float32]]
     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 +35,7 @@ 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(
@@ -153,12 +155,12 @@ class Episode:
         """Get the next extra features"""
         return self.all_extras[1:]
 
-    @cached_property
+    @property
     def n_agents(self):
         """The number of agents in the episode"""
         return self.all_extras[0].shape[0]
 
-    @cached_property
+    @property
     def n_actions(self):
         """The number of actions"""
         return len(self.all_available_actions[0][0])
@@ -267,7 +269,7 @@ class Episode:
     def __len__(self):
         return self.episode_len
 
-    @cached_property
+    @property
     def score(self) -> list[float]:
         """The episode score (sum of all rewards across all objectives)"""
         score = []
@@ -363,51 +365,6 @@ 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/utils/__init__.py

@@ -1,5 +1,5 @@
-from .schedule import Schedule, MultiSchedule, RoundedSchedule, LinearSchedule, ExpSchedule
-
+from .cached_property_collector import CachedPropertyCollector, CachedPropertyInvalidator
+from .schedule import ExpSchedule, LinearSchedule, MultiSchedule, RoundedSchedule, Schedule
 
 __all__ = [
     "Schedule",
@@ -7,4 +7,6 @@ __all__ = [
     "ExpSchedule",
     "MultiSchedule",
     "RoundedSchedule",
+    "CachedPropertyCollector",
+    "CachedPropertyInvalidator",
 ]

marlenv/utils/cached_property_collector.py

@@ -0,0 +1,17 @@
+from functools import cached_property
+
+
+class CachedPropertyCollector(type):
+    def __init__(cls, name: str, bases: tuple, namespace: dict):
+        super().__init__(name, bases, namespace)
+        cls.CACHED_PROPERTY_NAMES = [key for key, value in namespace.items() if isinstance(value, cached_property)]
+
+
+class CachedPropertyInvalidator(metaclass=CachedPropertyCollector):
+    def __init__(self):
+        super().__init__()
+
+    def invalidate_cached_properties(self):
+        for key in self.__class__.CACHED_PROPERTY_NAMES:
+            if hasattr(self, key):
+                delattr(self, key)

marlenv/wrappers/__init__.py

@@ -11,6 +11,7 @@ from .centralised import Centralized
 from .available_actions_mask import AvailableActionsMask
 from .delayed_rewards import DelayedReward
 from .potential_shaping import PotentialShaping
+from .action_randomizer import ActionRandomizer
 
 __all__ = [
     "RLEnvWrapper",
@@ -28,4 +29,5 @@ __all__ = [
     "Centralized",
     "DelayedReward",
     "PotentialShaping",
+    "ActionRandomizer",
 ]

marlenv/wrappers/action_randomizer.py

@@ -0,0 +1,17 @@
+from .rlenv_wrapper import RLEnvWrapper, AS, MARLEnv
+import numpy as np
+
+
+class ActionRandomizer(RLEnvWrapper[AS]):
+    def __init__(self, env: MARLEnv[AS], p: float):
+        super().__init__(env)
+        self.p = p
+
+    def step(self, action):
+        if np.random.rand() < self.p:
+            action = self.action_space.sample()
+        return super().step(action)
+
+    def seed(self, seed_value: int):
+        np.random.seed(seed_value)
+        super().seed(seed_value)

{multi_agent_rlenv-3.5.2.dist-info → multi_agent_rlenv-3.5.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.5.2
+Version: 3.5.5
 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-3.5.2.dist-info → multi_agent_rlenv-3.5.5.dist-info}/RECORD

@@ -1,4 +1,4 @@
-marlenv/__init__.py,sha256=UoZATsYMuKlnHyYdIRX7eQ6mGcmMww-tqX3uCyWVqRA,3656
+marlenv/__init__.py,sha256=bX76JknjwfVJ6IOKql_y4rIqvvx9raepD7u2lB9CgGo,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
@@ -12,15 +12,17 @@ marlenv/adapters/pymarl_adapter.py,sha256=2s7EY31s1hrml3q-BBaXo_eDMXTjkebozZPvzs
 marlenv/adapters/smac_adapter.py,sha256=8uWC7YKsaSXeTS8AUhpGOKvrWMbVEQT2-pml5BaFUB0,8343
 marlenv/models/__init__.py,sha256=uihmRs71Gg5z7Bvau_xtaQVg7xEtX8sTzi74bIHL5P0,443
 marlenv/models/env.py,sha256=BG1iVHxGD_p827mF0ewyOBn6wU2gtFsHLW1b4UtW-V0,7841
-marlenv/models/episode.py,sha256=IKPLuDVlz85Be6zrC21gyautjqRkEApS4fgRqQR52s0,15190
-marlenv/models/observation.py,sha256=kAmh1hIoC2TGrZlGVzV0y4TXXCSrI7gcmG0raeoncYk,3153
+marlenv/models/episode.py,sha256=zsyxsW4LIioPKyY4DZKn64A31e5ZvlwOf3HIGuRUzhs,13531
+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
+marlenv/utils/__init__.py,sha256=36pNw0r4V3xsqPZ5ljM29o96dfPAFq8WvMwggyv41fI,362
+marlenv/utils/cached_property_collector.py,sha256=IOjbr61f0DqLhcidXKrl7MhN1BOEGiTzCANIKQCxaF0,600
 marlenv/utils/schedule.py,sha256=slhtpQiBHSUNyPmSkKb2yBgiHJqPhoPxa33GxvyV8Jc,8565
-marlenv/wrappers/__init__.py,sha256=uV00m0jysZBgOW-TvRekis-gsAKPeR51P3HsuRZKxG8,880
+marlenv/wrappers/__init__.py,sha256=Z4_M-mxRNKQeu52tkmQ4B2m3-zrsmjfXXL5NsWQ4vu4,952
+marlenv/wrappers/action_randomizer.py,sha256=A1kejqGOTA0sc_RQL0EOd6sMSbcIdiV5zlscjKUlzdY,474
 marlenv/wrappers/agent_id_wrapper.py,sha256=9qHV3LMQ4AjcDCSuvQhz5h9hUf7Xtrdi2sIxmNZk5NA,1126
 marlenv/wrappers/available_actions_mask.py,sha256=OMyt2KntsR8JA2RuRgvwdzqzPe-_H-KKkbUUJfe_mks,1404
 marlenv/wrappers/available_actions_wrapper.py,sha256=_HRl9zsjJgSrLgVuT-BjpnnfrfM8ic6wBUWlg67uCx4,926
@@ -34,7 +36,7 @@ marlenv/wrappers/potential_shaping.py,sha256=T_QvnmWReCgpyoInxRw2UXbmdvcBD5U-vV1
 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-3.5.2.dist-info/METADATA,sha256=QjQkN0ZJsbaa-GyP7fAs4JFSTJkEUBLrIV0zCGPUvrc,4897
-multi_agent_rlenv-3.5.2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-multi_agent_rlenv-3.5.2.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
-multi_agent_rlenv-3.5.2.dist-info/RECORD,,
+multi_agent_rlenv-3.5.5.dist-info/METADATA,sha256=WKf56Bb7PqZFrw2B6Sx8zulM-h7aZRqXMTvYHrSxEtQ,5005
+multi_agent_rlenv-3.5.5.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+multi_agent_rlenv-3.5.5.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
+multi_agent_rlenv-3.5.5.dist-info/RECORD,,