更改继承类的__init__方法中的变量

Question

在构建从其他类继承的类的实例后，我试图更改

seed

方法内的“

__init__

”的默认值。我在“lab2d/dmlab2d/__init__.py”中有

“环境”类

。当我打印我的类的这个参数

print(f"dmlab2d  {env._env.__class__.__bases__[0].__bases__[0].__dict__}")

时，输出是

dmlab2d {'__module__': 'dmlab2d', '__doc__': 'Environment class for DeepMind Lab2D.\n\n  This environment extends the `dm_env` interface with additional methods.\n  For details, see https://github.com/deepmind/dm_env\n  ', '__init__': <function Environment.__init__ at 0x78d207cfd5a0>, 'reset': <function Environment.reset at 0x78d207cfd510>, '_read_action': <function Environment._read_action at 0x78d207cfd480>, 'step': <function Environment.step at 0x78d207cfd120>, 'observation': <function Environment.observation at 0x78d207cfd360>, 'observation_spec': <function Environment.observation_spec at 0x78d207cfd2d0>, '_make_observation_spec': <function Environment._make_observation_spec at 0x78d207cfd240>, '_make_action_spec': <function Environment._make_action_spec at 0x78d207cfd090>, 'action_spec': <function Environment.action_spec at 0x78d207cfc310>, 'events': <function Environment.events at 0x78d207cfc280>, 'list_property': <function Environment.list_property at 0x78d207cfc1f0>, 'write_property': <function Environment.write_property at 0x78d207cfc0d0>, 'read_property': <function Environment.read_property at 0x78d207cfc040>, '__abstractmethods__': frozenset(), '_abc_impl': <_abc._abc_data object at 0x78d207b66400>}

有人可以建议我如何在固有类属性的这种纠结情况下更改

seed

类中的

Environment

值吗？

这是我的代码

import dmlab2d
import gymnasium as gym
from matplotlib import pyplot as plt
from gymnasium import spaces
from meltingpot import substrate

from ml_collections import config_dict
import numpy as np
from ray.rllib.env import multi_agent_env
class MeltingPotEnv(multi_agent_env.MultiAgentEnv):
  """An adapter between the Melting Pot substrates and RLLib MultiAgentEnv."""

  def __init__(self, env: dmlab2d.Environment, max_cycles: int = MAX_CYCLES):
    """Initializes the instance.

    Args:
      env: dmlab2d environment to wrap. Will be closed when this wrapper closes.
    """
    self._env = env
    self._num_players = len(self._env.observation_spec())
    self._ordered_agent_ids = [
        PLAYER_STR_FORMAT.format(index=index)
        for index in range(self._num_players)
    ]
    # RLLib requires environments to have the following member variables:
    # observation_space, action_space, and _agent_ids
    self._agent_ids = set(self._ordered_agent_ids)
    # RLLib expects a dictionary of agent_id to observation or action,
    # Melting Pot uses a tuple, so we convert
    self.observation_space = self._convert_spaces_tuple_to_dict(
        spec_to_space(self._env.observation_spec()),
        remove_world_observations=True)
    self.action_space = self._convert_spaces_tuple_to_dict(
        spec_to_space(self._env.action_spec()))
    self.max_cycles = max_cycles
    self.num_cycles = 0

    super().__init__()

  def reset(self, *args, **kwargs):
    """See base class."""
    timestep = self._env.reset()
    self.num_cycles = 0
    return timestep_to_observations(timestep), {}

  def step(self, action_dict):
    """See base class."""
    actions = [action_dict[agent_id] for agent_id in self._ordered_agent_ids]
    timestep = self._env.step(actions)
    rewards = {
        agent_id: timestep.reward[index]
        for index, agent_id in enumerate(self._ordered_agent_ids)
    }
    self.num_cycles += 1
    termination = timestep.last()
    done = { '__all__': termination}
    truncation = self.num_cycles >= self.max_cycles
    truncations = {agent_id: truncation for agent_id in self._ordered_agent_ids}
    info = {}

    observations = timestep_to_observations(timestep)
    return observations, rewards, done, truncations, info

 

  def get_dmlab2d_env(self):
    """Returns the underlying DM Lab2D environment."""
    return self._env

  # Metadata is required by the gym `Env` class that we are extending, to show
  # which modes the `render` method supports.
  metadata = {'render.modes': ['rgb_array']}

  def render(self) -> np.ndarray:
    """Render the environment.

    This allows you to set `record_env` in your training config, to record
    videos of gameplay.

    Returns:
        np.ndarray: This returns a numpy.ndarray with shape (x, y, 3),
        representing RGB values for an x-by-y pixel image, suitable for turning
        into a video.
    """
    observation = self._env.observation()
    world_rgb = observation[0]['WORLD.RGB']

    # RGB mode is used for recording videos
    return world_rgb

  def _convert_spaces_tuple_to_dict(
      self,
      input_tuple: spaces.Tuple,
      remove_world_observations: bool = False) -> spaces.Dict:
    """Returns spaces tuple converted to a dictionary.

    Args:
      input_tuple: tuple to convert.
      remove_world_observations: If True will remove non-player observations.
    """
    return spaces.Dict({
        agent_id: (remove_world_observations_from_space(input_tuple[i])
                   if remove_world_observations else input_tuple[i])
        for i, agent_id in enumerate(self._ordered_agent_ids)
    })
env = substrate.build(env_config['substrate'], roles=env_config['roles'])
env = MeltingPotEnv(env)

Answer 1

seed

是用于此赋值的构造函数参数：

self._rng = np.random.RandomState(seed=seed)

如果要设置

seed

，可以通过构造函数参数来完成

Environment(env, observation_names, seed=your_seed)