multi-agent-rlenv 3.3.1__tar.gz → 3.3.2__tar.gz

{multi_agent_rlenv-3.3.1 → multi_agent_rlenv-3.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.3.1
+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>

{multi_agent_rlenv-3.3.1 → multi_agent_rlenv-3.3.2}/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__ = "3.3.1"
+__version__ = "3.3.2"
 
 from . import models
 from . import wrappers

{multi_agent_rlenv-3.3.1 → multi_agent_rlenv-3.3.2}/src/marlenv/adapters/overcooked_adapter.py

@@ -1,6 +1,7 @@
 import sys
 from dataclasses import dataclass
 from typing import Literal, Sequence
+from copy import deepcopy
 
 import cv2
 import numpy as np
@@ -87,6 +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():

{multi_agent_rlenv-3.3.1 → multi_agent_rlenv-3.3.2}/tests/test_serialization.py

@@ -1,15 +1,12 @@
 import pickle
-from importlib.util import find_spec
-
 import numpy as np
 import orjson
 import pytest
+from copy import deepcopy
 
 import marlenv
 from marlenv import DiscreteMockEnv
 
-skip_gym = find_spec("gymnasium") is None
-
 
 def test_registry():
     env = DiscreteMockEnv(4)
@@ -22,7 +19,7 @@ def test_registry():
     assert restored_env.n_actions == env.n_actions
 
 
-@pytest.mark.skipif(skip_gym, reason="Gymnasium is not installed")
+@pytest.mark.skipif(not marlenv.adapters.HAS_GYM, reason="Gymnasium is not installed")
 def test_registry_gym():
     env = marlenv.make("CartPole-v1")
     restored_env = pickle.loads(pickle.dumps(env))
@@ -135,3 +132,27 @@ def test_serialize_episode():
         episode.add(transition)
 
     _ = orjson.dumps(episode, option=orjson.OPT_SERIALIZE_NUMPY)
+
+
+@pytest.mark.skipif(not marlenv.adapters.HAS_OVERCOOKED, reason="Overcooked is not installed")
+def test_deepcopy_overcooked():
+    env = marlenv.adapters.Overcooked.from_layout("scenario4")
+    env2 = deepcopy(env)
+    assert env == env2
+
+
+@pytest.mark.skipif(not marlenv.adapters.HAS_OVERCOOKED, reason="Overcooked is not installed")
+def test_pickle_overcooked():
+    env = marlenv.adapters.Overcooked.from_layout("scenario1_s", horizon=60)
+    serialized = pickle.dumps(env)
+    restored = pickle.loads(serialized)
+    assert env == restored
+
+    env.reset()
+    restored.reset()
+
+    for _ in range(50):
+        actions = env.sample_action()
+        step = env.step(actions)
+        step_restored = restored.step(actions)
+        assert step == step_restored