85aea8b77a57afbb8d81a1235382b73bff6552be,examples/mujoco_all_sac_remote.py,,run_experiment,#Any#,96

Before Change

def run_experiment(variant):
    if variant["env_name"] == "humanoid-rllab":
        from rllab.envs.mujoco.humanoid_env import HumanoidEnv
        env = normalize(HumanoidEnv())
    elif variant["env_name"] == "swimmer-rllab":
        from rllab.envs.mujoco.swimmer_env import SwimmerEnv
        env = normalize(SwimmerEnv())
    else:
        env = normalize(GymEnv(variant["env_name"]))
    env = DelayedEnv(env, delay=0.01)

    pool = SimpleReplayPool(
        env_spec=env.spec,
        max_size=variant["max_pool_size"],
    )

    sampler = RemoteSampler(
        max_path_length=variant["max_path_length"],
        min_pool_size=variant["max_path_length"],
        batch_size=variant["batch_size"]
    )

    base_kwargs = dict(
        sampler=sampler,
        epoch_length=variant["epoch_length"],
        n_epochs=variant["n_epochs"],
        n_train_repeat=variant["n_train_repeat"],
        eval_render=False,
        eval_n_episodes=1,
        eval_deterministic=True,
    )

    M = variant["layer_size"]
    qf = NNQFunction(
        env_spec=env.spec,
        hidden_layer_sizes=[M, M],
    )

    vf = NNVFunction(
        env_spec=env.spec,
        hidden_layer_sizes=[M, M],
    )

After Change

def run_experiment(variant):
    universe = variant["universe"]
    task = variant["task"]
    domain = variant["domain"]

    env = get_environment(universe, domain, task, env_params={})
    env = DelayedEnv(env, delay=0.01)

    pool = SimpleReplayPool(
        observation_shape=env.observation_space.shape,
        action_shape=env.action_space.shape,
        max_size=variant["max_pool_size"],
    )

    sampler = RemoteSampler(
        max_path_length=variant["max_path_length"],
        min_pool_size=variant["max_path_length"],
        batch_size=variant["batch_size"]
    )

    base_kwargs = dict(
        sampler=sampler,
        epoch_length=variant["epoch_length"],
        n_epochs=variant["n_epochs"],
        n_train_repeat=variant["n_train_repeat"],
        eval_render=False,
        eval_n_episodes=1,
        eval_deterministic=True,
    )

    M = variant["layer_size"]
    qf = NNQFunction(
        observation_shape=env.observation_space.shape,
        action_shape=env.action_space.shape,
        hidden_layer_sizes=[M, M],
    )

    vf = NNVFunction(
        observation_shape=env.observation_space.shape,
        hidden_layer_sizes=[M, M],
    )

    policy = GMMPolicy(
        observation_shape=env.observation_space.shape,
        action_shape=env.action_space.shape,
        K=variant["K"],
        hidden_layer_sizes=[M, M],

Instances