multi-agent-rlenv 3.5.4__tar.gz → 3.5.5__tar.gz

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.5.4
+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>

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/src/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__ = "3.5.4"
+__version__ = "3.5.5"
 
 from . import models
 from .models import (

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/src/marlenv/models/episode.py

@@ -2,20 +2,22 @@ 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]]
@@ -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 = []

multi_agent_rlenv-3.5.5/src/marlenv/utils/__init__.py

@@ -0,0 +1,12 @@
+from .cached_property_collector import CachedPropertyCollector, CachedPropertyInvalidator
+from .schedule import ExpSchedule, LinearSchedule, MultiSchedule, RoundedSchedule, Schedule
+
+__all__ = [
+    "Schedule",
+    "LinearSchedule",
+    "ExpSchedule",
+    "MultiSchedule",
+    "RoundedSchedule",
+    "CachedPropertyCollector",
+    "CachedPropertyInvalidator",
+]

multi_agent_rlenv-3.5.5/src/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)

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/src/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",
 ]

multi_agent_rlenv-3.5.5/src/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-3.5.4 → multi_agent_rlenv-3.5.5}/tests/test_episode.py

@@ -1,7 +1,8 @@
 import numpy as np
-from marlenv.models import Transition, Episode
+
+from marlenv import DiscreteMockEnv, Episode, Transition, wrappers
 from marlenv.exceptions import EnvironmentMismatchException, ReplayMismatchException
-from marlenv import wrappers, DiscreteMockEnv
+
 from .utils import generate_episode
 
 
@@ -219,3 +220,31 @@ def test_get_images():
     images = episode.get_images(env, seed=0)
     # +1 because there is one image for the initial state (after reset)
     assert len(images) == len(episode) + 1
+
+
+def test_episode_invalidate_cached_properties():
+    env = DiscreteMockEnv(end_game=10)
+    obs, state = env.reset()
+    episode = Episode.new(obs, state)
+    for _ in range(5):
+        action = env.sample_action()
+        step = env.step(action)
+        episode.add(Transition.from_step(obs, state, action, step))
+        obs = step.obs
+        state = step.state
+
+    # Access the cached properties
+    assert len(episode.states) == 5
+    assert len(episode.obs) == 5
+
+    # Invalidate the cached properties
+    episode.invalidate_cached_properties()
+    for _ in range(3):
+        action = env.sample_action()
+        step = env.step(action)
+        episode.add(Transition.from_step(obs, state, action, step))
+        obs = step.obs
+        state = step.state
+
+    # Access the cached properties again
+    assert len(episode.obs) == 8

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/tests/test_schedules.py

@@ -1,4 +1,4 @@
-from marlenv.utils import Schedule, MultiSchedule
+from marlenv.utils import MultiSchedule, Schedule
 
 
 def is_close(a: float, b: float, tol: float = 1e-6) -> bool:

{multi_agent_rlenv-3.5.4 → multi_agent_rlenv-3.5.5}/tests/test_wrappers.py

@@ -1,6 +1,6 @@
 import numpy as np
 from marlenv import Builder, DiscreteMOMockEnv, DiscreteMockEnv, MARLEnv
-from marlenv.wrappers import Centralized, AvailableActionsMask, TimeLimit, LastAction, DelayedReward
+from marlenv.wrappers import Centralized, AvailableActionsMask, TimeLimit, LastAction, DelayedReward, ActionRandomizer
 import marlenv
 
 

multi_agent_rlenv-3.5.4/src/marlenv/utils/__init__.py

@@ -1,10 +0,0 @@
-from .schedule import Schedule, MultiSchedule, RoundedSchedule, LinearSchedule, ExpSchedule
-
-
-__all__ = [
-    "Schedule",
-    "LinearSchedule",
-    "ExpSchedule",
-    "MultiSchedule",
-    "RoundedSchedule",
-]