multi-agent-rlenv 3.5.0__py3-none-any.whl → 3.5.2__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__ = "3.5.0"
+__version__ = "3.5.2"
 
 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

@@ -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,

marlenv/models/step.py

@@ -39,7 +39,7 @@ class Step:
             case int() | float():
                 self.reward = np.array([reward], dtype=np.float32)
             case np.ndarray():
-                self.reward = reward
+                self.reward = reward.astype(np.float32)
             case other:
                 # We assume this is a sequence of some sort
                 self.reward = np.array(other, dtype=np.float32)

marlenv/wrappers/__init__.py

@@ -10,6 +10,7 @@ from .blind_wrapper import Blind
 from .centralised import Centralized
 from .available_actions_mask import AvailableActionsMask
 from .delayed_rewards import DelayedReward
+from .potential_shaping import PotentialShaping
 
 __all__ = [
     "RLEnvWrapper",
@@ -26,4 +27,5 @@ __all__ = [
     "Blind",
     "Centralized",
     "DelayedReward",
+    "PotentialShaping",
 ]

marlenv/wrappers/potential_shaping.py

@@ -0,0 +1,54 @@
+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).
+
+    https://people.eecs.berkeley.edu/~pabbeel/cs287-fa09/readings/NgHaradaRussell-shaping-ICML1999.pdf
+    """
+
+    gamma: float
+
+    def __init__(
+        self,
+        env: MARLEnv,
+        gamma: float = 1.0,
+        extra_shape: Optional[tuple[int]] = None,
+    ):
+        super().__init__(env, extra_shape=extra_shape)
+        self.gamma = gamma
+        self._current_potential = self.compute_potential()
+
+    def add_extras(self, obs: Observation) -> Observation:
+        """Add the extras related to potential shaping. Does nothing by default."""
+        return obs
+
+    def reset(self):
+        obs, state = super().reset()
+        self._current_potential = self.compute_potential()
+        return self.add_extras(obs), state
+
+    def step(self, actions):
+        prev_potential = self._current_potential
+        step = super().step(actions)
+
+        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 | npt.NDArray[np.float32]:
+        """Compute the potential of the current state of the environment."""

marlenv/wrappers/rlenv_wrapper.py

@@ -1,7 +1,6 @@
 from typing import Optional, Sequence
 from typing_extensions import TypeVar
 from dataclasses import dataclass
-from abc import ABC
 import numpy as np
 
 from marlenv.models import MARLEnv, Space, DiscreteSpace, State
@@ -11,7 +10,7 @@ AS = TypeVar("AS", bound=Space, default=Space)
 
 
 @dataclass
-class RLEnvWrapper(MARLEnv[AS], ABC):
+class RLEnvWrapper(MARLEnv[AS]):
     """Parent class for all RLEnv wrappers"""
 
     wrapped: MARLEnv[AS]

{multi_agent_rlenv-3.5.0.dist-info → multi_agent_rlenv-3.5.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: multi-agent-rlenv
-Version: 3.5.0
+Version: 3.5.2
 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-3.5.0.dist-info → multi_agent_rlenv-3.5.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-marlenv/__init__.py,sha256=PDJGtS3Psbiif5tc-PyUCfIju4pSlc6ukb-if6rbOv0,3656
+marlenv/__init__.py,sha256=UoZATsYMuKlnHyYdIRX7eQ6mGcmMww-tqX3uCyWVqRA,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
@@ -11,16 +11,16 @@ 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/env.py,sha256=BG1iVHxGD_p827mF0ewyOBn6wU2gtFsHLW1b4UtW-V0,7841
+marlenv/models/episode.py,sha256=IKPLuDVlz85Be6zrC21gyautjqRkEApS4fgRqQR52s0,15190
 marlenv/models/observation.py,sha256=kAmh1hIoC2TGrZlGVzV0y4TXXCSrI7gcmG0raeoncYk,3153
 marlenv/models/spaces.py,sha256=v7jnhPfj7vq7DFFJpSbQEYe4NGLLlj_bj2pzvvSBX7Y,7777
 marlenv/models/state.py,sha256=958PXTHadi3gtRnhGgcGtqBnF44R11kdcx62NN2gwxA,1717
-marlenv/models/step.py,sha256=LKGAV2Cu-k9Gz1hwrfvGx51l8axtQRqDE9WVL5r2A1Q,3037
+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/schedule.py,sha256=slhtpQiBHSUNyPmSkKb2yBgiHJqPhoPxa33GxvyV8Jc,8565
-marlenv/wrappers/__init__.py,sha256=wl23NUYcl0vPJb2QLpe4Xj8ZocUIOarAZX8CgWqdSQE,808
+marlenv/wrappers/__init__.py,sha256=uV00m0jysZBgOW-TvRekis-gsAKPeR51P3HsuRZKxG8,880
 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
@@ -30,10 +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/rlenv_wrapper.py,sha256=9k0bH5oHfnxnQYXL5PIRVbJfiTP_jnh2c9stMC8bA1o,3018
+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-3.5.0.dist-info/METADATA,sha256=XiqIH0A9YzbwckC_1hf8KbOI99LRDyYY--rRqAoMxZE,4897
-multi_agent_rlenv-3.5.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-multi_agent_rlenv-3.5.0.dist-info/licenses/LICENSE,sha256=_eeiGVoIJ7kYt6l1zbIvSBQppTnw0mjnYk1lQ4FxEjE,1074
-multi_agent_rlenv-3.5.0.dist-info/RECORD,,
+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,,