def set_reward_from_config(self, env_name: str, env_config: dict):
"""Build and set the reward function from environment configurations.
Args:
env_name: the environment's id
env_config: the environment's configuration
"""
self.reward_fn = get_reward_fn(env_name, env_config)
def env_reward(request, envs, env_config):
env_name = request.param
if "HalfCheetah" in env_name:
env_config["exclude_current_positions_from_observation"] = False
if "IndustrialBenchmark" in env_name:
env_config["max_episode_steps"] = 200
env = envs[env_name](env_config)
reward_fn = get_reward_fn(env_name, env_config)
return env, reward_fn
def env_reward(request, envs, env_config):
env_name: str = request.param
if env_name.endswith("-v3"):
kwargs = dict(exclude_current_positions_from_observation=False)
env_config["kwargs"] = kwargs
if "IndustrialBenchmark" in env_name:
env_config["max_episode_steps"] = 200
env = envs[env_name](env_config)
reward_fn = get_reward_fn(env_name, env_config)
return env, reward_fn
def env_reward(request, envs, env_config):
env_name = request.param
if any(
prefix in env_name for prefix in "HalfCheetah Walker2d Swimmer Hopper".split()
):
env_config["exclude_current_positions_from_observation"] = False
if "IndustrialBenchmark" in env_name:
env_config["max_episode_steps"] = 200
env = envs[env_name](env_config)
reward_fn = get_reward_fn(env_name, env_config)
return env, reward_fn
def test_set_reward_from_config(policy, envs): # pylint:disable=unused-argument
obs_space, action_space = policy.observation_space, policy.action_space
batch_size = 10
obs = fake_space_samples(obs_space, batch_size=batch_size)
act = fake_space_samples(action_space, batch_size=batch_size)
new_obs = fake_space_samples(obs_space, batch_size=batch_size)
obs, act, new_obs = map(policy.convert_to_tensor, (obs, act, new_obs))
policy.set_reward_from_config("MockEnv", {})
original_fn = get_reward_fn("MockEnv", {})
expected_rew = original_fn(obs, act, new_obs)
rew = policy.reward_fn(obs, act, new_obs)
assert torch.allclose(rew, expected_rew)
def set_reward_from_config(self):
"""Build and set the reward function from environment configurations."""
env_id, env_config = self.config["env"], self.config["env_config"]
self.reward_fn = envs.get_reward_fn(env_id, env_config)
self._set_reward_hook()
def reward_fn(env_name, env_config, size):
base = envs.get_reward_fn(env_name, env_config)
wrapped = LinearRedundant.wrap_env_function(base, size)
return wrapped
def env_reward(request, envs, env_config):
env_name = request.param
env = envs[env_name](env_config)
reward_fn = get_reward_fn(env_name, env_config)
return env, reward_fn
def reward_fn(env_name, env_config, size):
base = envs.get_reward_fn(env_name, env_config)
wrapped = CorrelatedIrrelevant.wrap_env_function(base, size)
return wrapped
def reward_fn(env_name, size, env_config):
base = envs.get_reward_fn(env_name, env_config)
wrapped = RandomIrrelevant.wrap_env_function(base, size)
return wrapped